See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/264871899

Lecture Notes Introduction to Differential Geometry MATH 442

Article · January 2005

CITATIONS READS

0 9,078

1 author:

Ivan Avramidi
New Mexico Institute of Mining and Technology
136 PUBLICATIONS 2,739 CITATIONS

SEE PROFILE

All content following this page was uploaded by Ivan Avramidi on 01 September 2014.

The user has requested enhancement of the downloaded file.

 Lecture Notes
Introduction to Differential Geometry
MATH 442
Instructor: Ivan Avramidi

New Mexico Institute of Mining and Technology

Socorro, NM 87801

August 25, 2005

Author: Ivan Avramidi; File: diffgeom.tex; Date: April 12, 2006; Time: 17:59
Contents

1 Manifolds 1
1.1 Submanifolds of Euclidean Space . . . . . . . . . . . . . . . . . 1
1.1.1 Submanifolds of Rn . . . . . . . . . . . . . . . . . . . . . 2
1.1.2 Differential of a Map . . . . . . . . . . . . . . . . . . . . 4
1.1.3 Main Theorem on Submanifolds of Rm . . . . . . . . . . 5
1.2 Manifolds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Basic Notions of Topology . . . . . . . . . . . . . . . . . 6
1.2.2 Idea of a Manifold . . . . . . . . . . . . . . . . . . . . . 9
1.2.3 Rigorous Definition of a Manifold . . . . . . . . . . . . . 11
1.2.4 Complex Manifolds . . . . . . . . . . . . . . . . . . . . 12
1.3 Tangent Vectors and Mappings . . . . . . . . . . . . . . . . . . . 14
1.3.1 Tangent Vectors . . . . . . . . . . . . . . . . . . . . . . . 14
1.3.2 Vectors as Differential Operators . . . . . . . . . . . . . . 15
1.3.3 Tangent Space . . . . . . . . . . . . . . . . . . . . . . . 16
1.3.4 Mappings . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.5 Submanifolds . . . . . . . . . . . . . . . . . . . . . . . . 18
1.3.6 Change of Coordinates . . . . . . . . . . . . . . . . . . . 20
1.4 Vector Fields and Flows . . . . . . . . . . . . . . . . . . . . . . . 21
1.4.1 Vector Fields in Rn . . . . . . . . . . . . . . . . . . . . . 21
1.4.2 Vector Fields on Manifolds . . . . . . . . . . . . . . . . . 23
1.4.3 Straightening Flows . . . . . . . . . . . . . . . . . . . . 23

2 Tensors 25
2.1 Covectors and Riemannian Metric . . . . . . . . . . . . . . . . . 25
2.1.1 Linear Functionals and Dual Space . . . . . . . . . . . . 25
2.1.2 Differential of a Function . . . . . . . . . . . . . . . . . . 27
2.1.3 Inner Product . . . . . . . . . . . . . . . . . . . . . . . . 28
2.1.4 Riemannian Manifolds . . . . . . . . . . . . . . . . . . . 31

I
II CONTENTS

2.1.5 Curves of Steepest Ascent . . . . . . . . . . . . . . . . . 32

2.2 Tangent Bundle . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.1 Fiber Bundles . . . . . . . . . . . . . . . . . . . . . . . . 33
2.2.2 Tangent Bundle . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 The Cotangent Bundle . . . . . . . . . . . . . . . . . . . . . . . 38
2.3.1 Pull-Back of a Covector . . . . . . . . . . . . . . . . . . 39
2.3.2 Phase Space . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.3.3 The Poincaré 1-Form . . . . . . . . . . . . . . . . . . . . 42
2.4 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.4.1 Covariant Tensors . . . . . . . . . . . . . . . . . . . . . 44
2.4.2 Contravariant Tensors . . . . . . . . . . . . . . . . . . . 46
2.4.3 General Tensors of Type (p, q) . . . . . . . . . . . . . . . 48
2.4.4 Linear Transformations and Tensors . . . . . . . . . . . . 50
2.4.5 Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . . 51
2.4.6 Tensor Bundles . . . . . . . . . . . . . . . . . . . . . . . 51
2.4.7 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.4.8 Einstein Summation Convention . . . . . . . . . . . . . . 53

3 Differential Forms 55
3.1 Exterior Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.1.1 Permutation Group . . . . . . . . . . . . . . . . . . . . . 55
3.1.2 Permutations of Tensors . . . . . . . . . . . . . . . . . . 57
3.1.3 Alternating Tensors . . . . . . . . . . . . . . . . . . . . . 58
3.1.4 Exterior p-forms . . . . . . . . . . . . . . . . . . . . . . 62
3.1.5 Exterior Product . . . . . . . . . . . . . . . . . . . . . . 64
3.1.6 Interior Product . . . . . . . . . . . . . . . . . . . . . . . 66
3.2 Orientation and the Volume Form . . . . . . . . . . . . . . . . . 68
3.2.1 Orientation of a Vector Space . . . . . . . . . . . . . . . 68
3.2.2 Orientation of a Manifold . . . . . . . . . . . . . . . . . 69
3.2.3 Hypersurfaces in Orientable Manifolds . . . . . . . . . . 70
3.2.4 Projective Spaces . . . . . . . . . . . . . . . . . . . . . . 71
3.2.5 Pseudotensors and Tensor Densities . . . . . . . . . . . . 73
3.2.6 Volume Form . . . . . . . . . . . . . . . . . . . . . . . . 75
3.2.7 Star Operator and Duality . . . . . . . . . . . . . . . . . 77
3.3 Exterior Derivative . . . . . . . . . . . . . . . . . . . . . . . . . 80
3.3.1 Coderivative . . . . . . . . . . . . . . . . . . . . . . . . 82
3.4 Pullback of Forms . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.5 Vector Analysis in R3 . . . . . . . . . . . . . . . . . . . . . . . . 85

diffgeom.tex; April 12, 2006; 17:59; p. 1

CONTENTS III

3.5.1 Vector Algebra in R3 . . . . . . . . . . . . . . . . . . . . 85

3.5.2 Vector Analysis in R3 . . . . . . . . . . . . . . . . . . . . 86

4 Integration of Differential Forms 89

4.1 Integration over a Parametrized Subset . . . . . . . . . . . . . . . 89
4.1.1 Integration of n-Forms in Rn . . . . . . . . . . . . . . . . 89
4.1.2 Integration over Parametrized Subsets . . . . . . . . . . . 90
4.1.3 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . 91
4.1.4 Surface Integrals . . . . . . . . . . . . . . . . . . . . . . 91
4.1.5 Independence of Parametrization . . . . . . . . . . . . . . 92
4.1.6 Integrals and Pullbacks . . . . . . . . . . . . . . . . . . . 93
4.2 Integration over Manifolds . . . . . . . . . . . . . . . . . . . . . 95
4.2.1 Partition of Unity . . . . . . . . . . . . . . . . . . . . . . 95
4.2.2 Integration over Submanifolds . . . . . . . . . . . . . . . 97
4.2.3 Manifolds with boundary . . . . . . . . . . . . . . . . . . 98
4.3 Stokes’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.3.1 Orientation of the Boundary . . . . . . . . . . . . . . . . 99
4.3.2 Stokes’ Theorem . . . . . . . . . . . . . . . . . . . . . . 99
4.4 Poincaré Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . 104

5 Lie Derivative 107

5.1 Lie Derivative of a Vector Field . . . . . . . . . . . . . . . . . . . 107
5.1.1 Lie Bracket . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.1.2 Flow generated by the Lie Bracket . . . . . . . . . . . . . 110
5.2 Lie Derivative of Forms and Tensors . . . . . . . . . . . . . . . . 112
5.2.1 Properties of Lie Derivative . . . . . . . . . . . . . . . . 114
5.3 Frobenius Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.3.1 Distributions . . . . . . . . . . . . . . . . . . . . . . . . 119
5.3.2 Frobenius Theorem . . . . . . . . . . . . . . . . . . . . . 123
5.3.3 Foliations . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.4 Degree of a Map . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.4.1 Gauss-Bonnet Theorem . . . . . . . . . . . . . . . . . . 128
5.4.2 Laplacian . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.4.3 Brouwer Degree . . . . . . . . . . . . . . . . . . . . . . 133
5.4.4 Index of a Vector Field . . . . . . . . . . . . . . . . . . . 137
5.4.5 Linking Number . . . . . . . . . . . . . . . . . . . . . . 140

diffgeom.tex; April 12, 2006; 17:59; p. 2

IV CONTENTS

6 Connection and Curvature 143

6.1 Affine Connection . . . . . . . . . . . . . . . . . . . . . . . . . . 143
6.1.1 Covariant Derivative . . . . . . . . . . . . . . . . . . . . 143
6.1.2 Curvature, Torsion and Levi-Civita Connection . . . . . . 146
6.1.3 Parallel Transport . . . . . . . . . . . . . . . . . . . . . . 149
6.2 Tensor Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
6.2.1 Covariant Derivative of Tensors . . . . . . . . . . . . . . 150
6.2.2 Ricci Identities . . . . . . . . . . . . . . . . . . . . . . . 151
6.2.3 Normal Coordinates . . . . . . . . . . . . . . . . . . . . 152
6.2.4 Properties of the Curvature Tensor . . . . . . . . . . . . . 153
6.2.5 Bianci Identities . . . . . . . . . . . . . . . . . . . . . . 155
6.3 Cartan’s Structural Equations . . . . . . . . . . . . . . . . . . . . 157

7 Homology Theory 163

7.1 Algebraic Preliminaries . . . . . . . . . . . . . . . . . . . . . . . 163
7.1.1 Groups . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
7.1.2 Finitely Generated and Free Abelian Groups . . . . . . . 167
7.1.3 Cyclic Groups . . . . . . . . . . . . . . . . . . . . . . . 167
7.2 Singular Chains . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
7.2.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 175
7.3 Singular Homology Groups . . . . . . . . . . . . . . . . . . . . . 176
7.3.1 Cycles, Boundaries and Homology Groups . . . . . . . . 176
7.3.2 Simplicial Homology . . . . . . . . . . . . . . . . . . . . 177
7.3.3 Betti Numbers and Topological Invariants . . . . . . . . . 178
7.3.4 Some Theorems from Algebraic Topology . . . . . . . . . 180
7.3.5 Examples. . . . . . . . . . . . . . . . . . . . . . . . . . . 184
7.4 de Rham Cohomology Groups . . . . . . . . . . . . . . . . . . . 186
7.4.1 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . 189
7.5 Harmonic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . 190
7.6 Relative Homology and Morse Theory . . . . . . . . . . . . . . . 197
7.6.1 Relative Homology . . . . . . . . . . . . . . . . . . . . . 197
7.6.2 Morse Theory . . . . . . . . . . . . . . . . . . . . . . . . 201

8 207
8.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208

Bibliography 209

diffgeom.tex; April 12, 2006; 17:59; p. 3

CONTENTS V

Notation 211

diffgeom.tex; April 12, 2006; 17:59; p. 4

VI CONTENTS

diffgeom.tex; April 12, 2006; 17:59; p. 5

Chapter 1

Manifolds

1.1 Submanifolds of Euclidean Space

• Idea: Manifold is a general space that looks locally like a Euclidean space
of the same dimension. This allows to develop the differential and integral
calculus.

• Let n ∈ N be a positive integer. The Euclidean space Rn is a set of points x

described by ordered n-tuples (x1 , . . . , xn ) or real numbers.

• The numbers xi ∈ R, i = 1, . . . , n, are called the Cartesian coordinates of

the point x.

• The integer n is the dimension of the Euclidean space.

• The distance between two points of the Euclidean space is defined by

v
t n
X
d(x, y) = (xk − yk )2 .
k=1

• The open ball of radius ε centered at x0 is the set of points defined by

Bε (x0 ) = {x ∈ Rn | d(x, x0 ) < ε} .

• A neighborhood of a point x0 is the set of points that contain an open ball

around it.

1
2 CHAPTER 1. MANIFOLDS

• Let x0 ∈ Rn be a fixed point with Cartesian coordinates xi , i = 1, . . . , n,

in the Euclidean space and S ⊂ Rn be a neighborhood of x0 . An injective
(one-to-one) map
f : S → Rn
defined by
yi = f i (x1 , . . . , xn ) , i = 1, . . . , n,
where f i (x) are smooth functions, is called a coordinate system in S .

• The map f is injective if for any point x in S

∂fi
!
det , 0.
∂x j

1.1.1 Submanifolds of Rn
• Let n, r ∈ N be positive integers and m = n + r. Let M ⊂ Rm be a subset of
the Euclidean space Rm .

• Then M is a submanifold of Rm of dimension n if for any point x0 ∈ M in M

there exists a neighborhood with a coordinate system such that every point
in this neighborhood has coordinates (x1 , . . . , xn , xn+1 , . . . , xn+r ), where the
last r coordinates (xn+1 , . . . , xn+r ) are given by smooth functions of the first
n coordinates (x1 , . . . , xn ):

xα = f α (x1 , . . . , xn ) , α = n + 1, . . . , n + r .

• The coordinates (x1 , . . . , xn ) are called the local coordinates for M near x0 .

• More generally, let

F : Rm → Rr
be a smooth map described by r equations

yα = F α (x1 , . . . , xm ) , α = 1, . . . , r .

Let y0 ∈ Rr and

M = F −1 (y0 ) = {x ∈ Rm | F(x) = y0 }

diffgeom.tex; April 12, 2006; 17:59; p. 6

1.1. SUBMANIFOLDS OF EUCLIDEAN SPACE 3

be a subset of the Euclidean space Rm described by the locus of r equations

F α (x1 , . . . , xm ) = yα0 , α = 1, . . . , r .
Suppose that M is non-empty and let x0 ∈ M, that is
F(x0 ) = y0 .

• Then the Implicit Function Theorem says that if

∂F α
!
det (x0 ) , 0 ,
∂xβ
where α = 1, . . . , r and β = n + 1, . . . , n + r, then there is a neighborhood of
x0 such that the last r coordinates can be expressed as smooth functions of
the first n coordinates:
xα = f α (x1 , . . . , xn ) , α = n + 1, . . . , n + r .

• If this is true for any point of M, then M is a n-dimensional submanifold of

Rm .
• The matrix
∂F α
!
,
∂x j
where α = 1, . . . , r and j = 1, . . . , m is called the Jacobian matrix.
• The General Implicit Function Theorem says that if at a point x0 the Ja-
cobian matrix has the maximal rank equal to r,
∂F α
!
rank (x0 ) = r ,
∂x j
then there exists a coordinate system in a neighborhood of x0 such that the
last r coordinates can be expressed as smooth functions of the first n coor-
dinates.
• If this is true for every point of M, then M is a n-dimensional submanifold
of Rm .
• The number r is called the codimension of M.
• If the codimension r is equal to 1, that is n = m − 1, then M is called a
hypersurface.

diffgeom.tex; April 12, 2006; 17:59; p. 7

4 CHAPTER 1. MANIFOLDS

1.1.2 Differential of a Map

• Let Rn be a Euclidean space and x0 ∈ Rm . Then the tangent space to Rm at
x0 is a vector space Rmx of all vectors in Rm based at x.

• Let
F : Rm → Rr

be a smooth map described by r smooth functions

yα = F α (x1 , . . . , xn ) , α = 1, . . . , r .

Let x0 ∈ Rm and x = x(t), t ∈ (−ε, ε), be a curve in Rm such that

dx
x(0) = x0 and (0) = v ,
dt

where v ∈ Rmx0 is the tangent vector to the curve at x0 .

• Let y0 = F(x0 ) ∈ Rr and y(t) = F(x(t)). Then

dy
y(0) = y0 and (0) = w ,
dt

where w ∈ Rrx0 is the tangent vector to the image of the curve at y0 .

• We compute
m
∂F α
X !
α
w = (x0 )vi .
i=1
∂x i

• Thus there is a linear transformation

F∗ : Rmx0 → Rry0 ,

so that
F∗ v = w .

F∗ is called the differential of the map F at x0 .

diffgeom.tex; April 12, 2006; 17:59; p. 8

1.1. SUBMANIFOLDS OF EUCLIDEAN SPACE 5

1.1.3 Main Theorem on Submanifolds of Rm

• The matrix of the linear transformation F∗ is exactly the Jacobian matrix.

• Therefore, the differential F∗ at a point x0 is a surjective (onto) map if and

only if m ≥ r and the Jacobian at x0 has the maximal rank equal to r.

• Recall that F∗ : Rmx0 → Rry0 is surjective if for any w ∈ Rry0 there is v ∈ Rmx0
such that F∗ v = w.

• Thus, we have the following theorem.

Theorem 1.1.1 Let F : Rm → Rr with m > r, y0 ∈ Rr and

M = F −1 (y0 ) = {x ∈ Rm | F(x) = y0 } .
•
If M is non-empty and for any x0 ∈ M the differential F∗ : Rmx0 → Rry0 is
surjective, then M is a n = (m − r)-dimensional submanifold of Rm .

diffgeom.tex; April 12, 2006; 17:59; p. 9

6 CHAPTER 1. MANIFOLDS

1.2 Manifolds
1.2.1 Basic Notions of Topology
• First we define the basic topological notions in the Euclidean space Rn .

• Let x0 ∈ Rn be a point in Rn and ε > 0 be a positive real number.

• The open ball in Rn of radius ε with the center at x0 is the set

B̄ε (x0 ) = {x ∈ Rn | d(x, x0 ) < ε} .

• The closed ball in Rn of radius ε with the center at x0 is the set

B̄ε (x0 ) = {x ∈ Rn | d(x, x0 ) ≤ ε} .

• Let U ⊆ Rn be a subset of Rn . A point x ∈ U is an interior point of U

if there is an open ball Bε (x) of some radius ε > 0 centered at x such that
Bε (x) ⊂ U.

• A point x ∈ Rn is a boundary point of U if every open ball Bε (x) of any

radius ε > 0 centered at x contains at least one point from U and one point
from its complement (that is not from U).

• The set U o of all interior points of U is called the interior of U.

• The set ∂U of all boundary points of U is called the boundary of U.

• A set U ⊆ Rn is open if every point of U is an interior point of U, that is,

U = U o.

• A set F ⊆ Rn is closed if its complement Rn \F is open, that is, F = F o ∪∂F.

• The sets ∅ and Rn are both open and closed.

• The union of any collection of open sets is open.

• The intersection of any finite number of open sets is open.

diffgeom.tex; April 12, 2006; 17:59; p. 10

1.2. MANIFOLDS 7

Definition 1.2.1 A general topological space is a set M together with

a collection of subsets of M, called open sets, that satisfy the following
properties

1. M and ∅ are open,

•
2. the intersection of any finite number of open sets is open,

3. the union of any collection of open sets is open.

Such a collection of open sets is called a topology of M.

• A subset of M is closed if its complement M \ F is open.

• The closure S̄ of a subset S ⊆∈ M of a topological space M is the intersec-

tion of all closed sets that contain S ; it is equal to S̄ = S ∪ ∂S .

• A subset S ⊆ M of a topological space is dense in M if S̄ = M, that is,

every non-empty subset of M contains an element of S .

• A topological space is called separable if it contains a countable dense

subset.

• A topology on M naturally induces a topology on any subset of M. Let

A ⊆ M be a subset of M. Then the induced topology on A is defined as
follows. A subset V ⊆ A is defined to be open subset of A if there is an open
subset U ⊆ M of M such that V = U ∩ A.

• Let x ∈ M be a point in a topological space M. An open set in M containing

the point x is called a neighborhood of x.

Definition 1.2.2 Let M and N be two topological spaces and F : M →

N be a map from M into N. The map F is said to be continuous if the
inverse image of any open set in N is an open set in M.

• That is, if for any open set V ⊂ N the set

F −1 (V) = {x ∈ M | F(x) ∈ V}

is open in M.

• The direct images of open sets do not have to be open!

diffgeom.tex; April 12, 2006; 17:59; p. 11

8 CHAPTER 1. MANIFOLDS

• Let the map F be bijective, that is, injective (one-to-one) and surjective
(onto). Then there exists the inverse map F −1 : N → M.

Definition 1.2.3 A map F : M → N is called a homeomorphism if it

•
is bijective and both F and F −1 are continuous.

• Homeomorphisms preserve topology, that is, they take open sets to open
sets and closed sets to closed sets.

• A topological space M is called Hausdorff if any two points of M have

disjoint neighborhoods.

• A collection of subsets of a topological space M is called an open cover of

M if the union of all subsets in the collection coincides with M.

• A subcollection of subsets that is itself a cover is called a subcover.

Definition 1.2.4 A topological space M is called compact if every
•
open cover of M has a finite subcover.

• A subset M ∈ Rn of a Euclidean space if called bounded if there is an open

ball BC (0) of some radius C centered at the origin such that M ⊆ BC (0).

Theorem 1.2.1 Bolzano-Weierstrass Theorem. Let M ⊆ Rn be a

• subset of Rn with the induced topology. Then M is compact if and only
if M is closed and bounded in Rn .

• Properties of Continuous Maps.

Theorem 1.2.2 Let M and N be two topological spaces and M be

compact. Let F : M → N be a continuous map from M into N. Then
• the image F(M) of M is compact in N.

That is, a continuous image of a compact topological space is compact.

Proof :

1. Let {Uα }α∈A be an open cover of F(M) in N.

2. Then {F −1 (Uα )}α∈A is an open cover of M.
3. Since M is compact it has a finite subcover {F −1 (Ui )}ni=1 .
4. Then {Ui }ni=1 is a finite subcover of F(M).

diffgeom.tex; April 12, 2006; 17:59; p. 12

1.2. MANIFOLDS 9

5. Thus F(M) is compact.



Theorem 1.2.3 A continuous real-valued function f : M → R on a

•
compact topological space M is bounded.
Proof :

1. f (M) is compact in Rn .
2. Thus f (M) is closed and bounded.


1.2.2 Idea of a Manifold

• A manifold M of dimension n is a topological space that is locally homeo-
morphic to Rn .

• A manifold M is covered by a family of local coordinate systems {Uα ; xα1 , . . . , xαn }α∈A ,
called an atlas, consisting of open sets, called patches (or charts), Uα , and
coordinates xα .

• A point p ∈ Uα ∩ Uβ that lies in two coordinate patches has two sets of

coordinates xα and xβ related by smooth functions

xαi = fαβ
i
(xβ1 , . . . , xβn ) , i = 1, . . . , n .

• The coordinates xα and xβ are said to be compatible.

• If all the functions fαβ are smooth, then the manifold M is said to be smooth.
If these functions are analytic, then the manifold is said to be real analytic.

• Each patch is homeomorphic to some open subset in Rn .

• Thus, a manifold is locally Euclidean.

• The collection of all patches (charts) is called an atlas.

• The collection of all coordinate systems that are compatible with those used
to define a manifold is called the maximal atlas.

diffgeom.tex; April 12, 2006; 17:59; p. 13

10 CHAPTER 1. MANIFOLDS

• Let M be an n-dimensional manifold with local coordinate systems {Uα ; xα1 , . . . , xαn }α∈A
and N be an m-dimensional manifold with local coordinate systems {Vβ ; y1β , . . . , ymβ }β∈B .
The product manifold L = M × N is a manifold

L = M × N = {(p, q) | p ∈ M, q ∈ N}

with local coordinate systems {Wαβ ; z1αβ , . . . , zn+m

αβ }α∈A,β∈B where

Wαβ = Uα × Vβ

and
(z1αβ , . . . , znαβ ) = (xα1 , . . . , xαn )
and
αβ , . . . , zαβ ) = (yβ , . . . , yβ ) .
(zn+1 n+m 1 m

Examples
• Unit Sphere S n .
S n = {x ∈ Rn+1 | d(x, 0) = 1}
Stereographic projection:

Φ : Rn+1 → Rn

Let v
n
t
X
R= 1− (x j )2
j=1

Then
x1 xn
!
x n+1
= ±R , Φ1,2 (x) = ,··· , .
1±R 1±R

• Torus T n = S 1 × · · · × S 1 .

– T n has local coordinates (θ1 , · · · , θn ) (angles).

– Topologically it is the cube [0, 1]n with the antipodal points identified.

• Real Projective Space RPn .

– RPn is the space of unoriented lines through the origin of Rn+1 .

diffgeom.tex; April 12, 2006; 17:59; p. 14

1.2. MANIFOLDS 11

– The set of oriented lines through the origin of Rn+1 is S n .

– Topologically RPn is the sphere S n in Rn+1 with the antipodal points
identified, which is the unit ball in Rn with the antipodal points on the
boundary (which is a unit sphere S n−1 ) identified.
– RPn is covered by (n + 1) sets

U j = {L ∈ RPn | L with x j , 0} , j = 1, . . . , n + 1 .

– The local coordinates in U j are

x1 xn
v1 = , · · · , vn
= .
xj xj

– The (n + 1)-tuple (x1 , . . . , xn+1 ) identified with (λx1 , . . . , λxn+1 ), λ , 0,

are called homogeneous coordinates of a point in RPn .

1.2.3 Rigorous Definition of a Manifold

• Let M be a set (without topology) and {Uα }α∈A be a collection of subsets
that is a cover of M, that is,
[
Uα = M .
α∈A

• Let
ϕα : Uα → Rn , α ∈ A,
be injective maps such that ϕα (Uα ) are open subsets in Rn .

• The set ϕα (Uα ∩ Uβ ) is an open subset in Rn . The maps

fαβ = ϕα ◦ ϕ−1
β : ϕα (U α ∩ U β ) → R

are called the transition functions (or the overlap functions). We assume
that the transition functions are smooth.

• The pair (Uα , ϕα ) is called a coordinate patch (or chart).

• A point p ∈ Uα is assigned coordinates of the point ϕα (p) in Rn . Thus, ϕα is

called a coordinate map.

diffgeom.tex; April 12, 2006; 17:59; p. 15

12 CHAPTER 1. MANIFOLDS

• Now, we take the maximal atlas of such coordinate patches.

• The topology in M is defined as follows.

• Let W ⊂ M be a subset of M. A point p ∈ W in W is said to be an interior

point if there is a coordinate chart (Uα , ϕα ) including p such that U ⊂ W.

• A subset W of M is declared to be open if all of its points are interior points.

• If the resulting topological space is Hausdorff and separable, then M is said

to be an n-dimensional smooth manifold.

• The regularity of the transition functions determines the regularity of the

manifold. If the transition functions are only differentiable once, then the
manifold is called differentiable. If they are of class C k , then the manifold
is called a manifold of class C k . If they are of class C ∞ , then the manifold
is called smooth. If the transition functions are analytic, then the manifold
is called analytic.

• Let F : M → R be a real-valued function on M. Let (Uα , xα ) be a local

coordinate system. Then the function

Fα = F ◦ ϕ−1
α : ϕα (U α ) → R

is a function of n real variables Fα (xα1 , . . . , xαn ).

• The function F is said to be smooth if the function Fα is smooth in terms

of local coordinates xα .

• The process of replacing the map F by the function Fα = F ◦ ϕ−1α is usually

omitted, and the functions F and Fα are identified, so that we think of the
function F directly in terms of local coordinates.

1.2.4 Complex Manifolds

• Let M be a set and {Uα }α∈A be its cover. Let

ϕα : Uα → Cn

be injective maps from Uα to the complex n-space Cn so that ϕα (Uα ) are

open subsets of Cn .

diffgeom.tex; April 12, 2006; 17:59; p. 16

1.2. MANIFOLDS 13

• Let the transition functions

fαβ = ϕα ◦ ϕ−1
β : ϕα (U α ∩ U β ) → C ,
n

defined by
zkα = fαβ
k
(z1β , . . . , znβ ) ,
where zkα = xαk + iykα and zkβ = xβk + iykβ , k = 1, . . . , n, be complex analytic.
That is, they satisfy Cauchy-Riemann conditions

∂xαk ∂ykα ∂xαk ∂ykα

= , =− ,
∂xβj ∂yβj ∂yβj ∂xβj
or
∂zkα
= 0,
∂z̄βj
with j, k = 1, . . . , n.

• Then M is called a n-dimensional complex manifold.

• The topological dimension of a n-dimensional complex manifold is 2n.

diffgeom.tex; April 12, 2006; 17:59; p. 17

14 CHAPTER 1. MANIFOLDS

1.3 Tangent Vectors and Mappings

• A tangent vector to a submanifold M of Rn at a point p0 ∈ M is the velocity
vector ṗ at p0 to some parametrized curve p = p(t) lying on M and passing
through the point p0 .

• Whitney theorem: Every n-dimensional manifold can be realized as a sub-

manifold of R2n , or as a smooth submanifold of R2n+1 .

• So, every manifold is a submanifold of a Euclidean space.

• However, the intrinsic geometry of M does not depend on its embedding in

a Euclidean space.

1.3.1 Tangent Vectors

• We fix a point p0 ∈ M on manifold M and consider a curve p : (−ε, ε) → M
such that
p(0) = p0 .

• Let (Uα , xαj ) be a local chart about p0 . The curve is described in local coor-
dinates by
xαi = xαi (t) .

• The velocity vector ṗ(0) is described in (Uα , xα ) by an n-tuple

 
ẋα1 (0), . . . , ẋαn (0) .

• Let (Uβ , xβj ) be another local chart containing p0 . Then the velocity vector
ṗ(0) is described in (Uβ , xβ ) by an n-tuple
 
ẋβ1 (0), . . . , ẋβn (0) .

• These n-tuples are related by the chain rule

 ∂xβ 

n
 i
X
ẋβi (0) =   (p0 ) ẋαj (0) .
j=1 ∂xα
j

diffgeom.tex; April 12, 2006; 17:59; p. 18

1.3. TANGENT VECTORS AND MAPPINGS 15

Definition 1.3.1 A tangent vector at a point p0 ∈ M of a manifold

M is a map that assigns to each coordinate chart (Uα , xα ) about p0 an
ordered n-tuple (Xα1 , . . . , Xαn ) such that
•
 ∂xβ 
n
 i
X
Xβ =
i   (p0 )Xαj .
j=1 ∂xα
j

1.3.2 Vectors as Differential Operators

• Let f : M → R be a real-valued function on M.

• Let p ∈ M be a point on M and X be a tangent vector at p.

• Let (Uα , xα ) be a coordinate chart about p.

• The (directional) derivative of f with respect to X at p (or along X, or in

the direction of X) is defined by
n
∂f
X !
X p ( f ) = DX ( f ) = (p)Xαj .
j=1 ∂xα
j

• Theorem 1.3.1 DX ( f ) does not depend on the local coordinate system.

Proof :

1. Chain rule.


• The intrinsic properties are invariant under a change of coordinate system.

They should not depend on the choice of the local chart.

• There is a one-to-one correspondence between tangent vectors to M at p and

first-order differential operators acting on real-valued functions in a local
coordinate chart (Uα , xα ) by
n
X ∂
Xp = Xαj .
j=1 ∂xαj p

• Therefore, we can identify tangent vectors and the differential operators.

diffgeom.tex; April 12, 2006; 17:59; p. 19

16 CHAPTER 1. MANIFOLDS

• Let us fix an index j = 1, . . . , n. The curve

xi (t) = x0i , i , j, x j (t) = t ,

is called the j-th coordinate curve.

• The velocity vector to this curve is given by

ẋi (0) = δij ,

where δij is the Kronecker symbol defined by

1, if i = j ,
(
δij =
0, if i , j .

• The differential operator corresponding to this velocity vector is

∂
.
∂xαj

1.3.3 Tangent Space

• Let M be a manifold and p ∈ M be a point in M. The tangent space T p M
to M at p is the real vector space of all tangent vectors to M at p.

• Let (U, x) be a local chart about p. Then the vectors

∂ ∂
,··· ,
∂x1 p
∂xn p

form a basis in the tangent space called the coordinate basis, or the coor-
dinate frame.

• A vector field X on an open set U ⊂ M in a manifold M is the differentiable

assignment of a tangent vector X p to each point p ∈ U.

• In local coordinates n
X ∂
X= X j (x) .
j=1
∂x j

• Example.

diffgeom.tex; April 12, 2006; 17:59; p. 20

1.3. TANGENT VECTORS AND MAPPINGS 17

1.3.4 Mappings

• Let M be an n-dimensional manifold and N be an m-dimensional manifold

and let F : M → N be a map from M to N. Let (Uα , ϕα )α∈A be an atlas in M
and (Vβ , ψβ )β∈B be an atlas in N. We define the maps

Fαβ = ψβ ◦ F ◦ ϕ−1
α : ϕα (U α ) → ψβ (Vβ )

from open sets in Rn to Rm defined by

yaβ = Fαβ
a
(xα1 , . . . , xαn ) ,

where a = 1, . . . , m.

a
• The map F is said to be smooth if Fαβ are smooth functions of local coor-
dinates xα , i = 1, . . . , n.
i

• The process of replacing the map F by the functions Fαβ = ψβ ◦ F ◦ ϕ−1 α is

usually omitted, and the maps F and Fαβ are identified, so that we think of
the map F directly in terms of local coordinates.

• If n = m and the map F : M → N is bijective and both F and F −1 are

differentiable, then F is called a diffeomorphism.

• That is, a diffeomorphism is a differentiable homeomorphism with differen-

tiable inverse.

• If this is only true in a neighborhood of a point p ∈ M, then F is called a

local diffeomorphism.

• Example.

diffgeom.tex; April 12, 2006; 17:59; p. 21

18 CHAPTER 1. MANIFOLDS

Definition 1.3.2 Let M and N be two manifolds and F : M → N be

a map from M into N. Let p0 ∈ M be a point in M and X ∈ T p0 M be
a tangent vector to M at p0 . Let p = p(t), t ∈ (−ε, ε), be a curve in M
such that
p(0) = p0 , ṗ(0) = X .

• Then the differential of F is the map

F∗ : T p0 M → T F(p0 ) N

defined by
d
F∗ X = F(p(t)) .
dt t=0

• F∗ does not depend on the curve p(t).

• The matrix of the linear transformation F∗ in therms of the coordinate bases

∂/∂yα and ∂/∂xi is the Jacobian matrix

α ∂yα
(F∗ ) i = i ,
∂x
that is,
n
α
X ∂yα
(F∗ X) i = Xi .
i=1
∂xi

1.3.5 Submanifolds
Definition 1.3.3 Let M be an n-dimensional manifold and W ⊂ M be
a subset of M. Then W is an r-dimensional embedded submanifold of
M if W is locally described as the common locus of (n − r) differentiable
independent functions

• F α (x1 , . . . , xn ) = 0 , α = 1, . . . , n − r ,

such that the Jacobian matrix has rank (n − r) at each point of the locus,
that is,
∂F α
!
rank (x) = n − r , ∀x ∈ W.
∂xi

• Every embedded submanifold of a manifold is itself a manifold.

diffgeom.tex; April 12, 2006; 17:59; p. 22

1.3. TANGENT VECTORS AND MAPPINGS 19

Theorem 1.3.2 Let n and r be two positive integers such that n > r.
Let M be an n-dimensional manifold and N be an r-dimensional man-
ifold. Let q ∈ N be a point in N such that the inverse image W =
F −1 (q) , ∅ is nonempty. Suppose that for each point p ∈ W the differ-
• ential F∗ of the map F is surjective, that is, has the maximal rank

rank F∗ (p) = r .

Then W is an (n − r) dimensional submanifold of M.

• Example. Morse Map. (Height function F : M → R for trousers surface

M in R3 ).

• Let p0 ∈ M and v ∈ T p0 M be a tangent vector to M at p0 . Then F∗ : T p0 M →

R = T F(p0 ) M is the projection of v to z-axis defined by F∗ (v1 , v2 , v3 ) = v3 .

• Let p0 ∈ M and z = F(p0 ). F −1 (z) is an embedded submanifold if F∗ is

onto, that is, T p0 M is not horizontal. If T p0 M is horizontal, then F∗ = 0
(hence, not onto).

Definition 1.3.4 Let M and N be two manifolds and F : M → N be a

differentiable map from M into N.

A point p ∈ M is a regular point if the differential F∗ : T p M → T F(p) N

is surjective.
•
A point p ∈ M is a critical point if it is not regular.

A point q ∈ N is a regular value of F if the inverse image F −1 (q) is

either empty or consists only of regular points.

A point q ∈ N is a critical value of F if it is not regular.

• Thus, we can reformulate the main theorem as follows.

Theorem 1.3.3 Let n and r be two positive integers such that n >
r. Let M be an n-dimensional manifold and N be an r-dimensional
manifold. Let q ∈ N be a regular value of F. Then the inverse image
W = F −1 (q) , ∅ is either empty or is an (n−r) dimensional submanifold
of M.

diffgeom.tex; April 12, 2006; 17:59; p. 23

20 CHAPTER 1. MANIFOLDS

Theorem 1.3.4 Sard’s Theorem. Let M and N be two manifolds and

F : M → N be a smooth map from M into N. Then the set of critical
•
values of F is a set of measure zero in N, that is, almost all values of F
are regular values.

• The critical values of a map F : M → N cannot fill up any open set in N.

1.3.6 Change of Coordinates

Theorem 1.3.5 Inverse Function Theorem. Let M and N be two n-
dimensional manifolds and F : M → N be a differentiable map from M
• into N. Let p0 ∈ M be such that the differential F∗ : T p0 M → T F(p0 ) N is
bijective. Then F is a local diffeomorphism in a neighborhood U of p0 ,
that is, F(U) is open in N and F : U → F(U) is a diffeomorphism.

Theorem 1.3.6 Let M be a manifold, p0 ∈ M be a point in M, and

(U, x) be a local coordinate chart about p0 . Let F : U → Rn be a
differentiable map defined by yi = F i (x), i = 1, . . . , n, such that

∂F i
!
•
det (p0 ) , 0 .
∂x j

Then there is a neighborhood V of p0 such that the n-tuple (y1 , . . . , yn )

forms a compatible coordinate system in V.

• Example.

diffgeom.tex; April 12, 2006; 17:59; p. 24

1.4. VECTOR FIELDS AND FLOWS 21

1.4 Vector Fields and Flows

1.4.1 Vector Fields in Rn
• Let x = (x j ) ∈ Rn be a point in Rn and v(x) ∈ T x Rn be a vector at x given by
n
X
v= v j (x)∂ j ,
j=1

where ∂ j = ∂/∂x j and the components v j (x) are smooth (or just differen-
tiable) functions of x. Then v(x) is called a vector field in Rn .

• Let t ∈ (−ε, ε) and

ϕ t : R n → Rn
be a family of diffeomorphisms such that

ϕ0 = Id

and for any t, t1 , t2 ∈ (−ε, ε) such that −t, t1 + t2 ∈ (−ε, ε) there holds

ϕt1 ◦ ϕt2 = ϕt1 +t2

and
ϕ−t = ϕ−1
t .

Such a one-parameter group of diffeomorphisms is called a flow on Rn .

• A flow ϕt defines a vector field v by

d
v(x) = ϕt (x)
dt t=0

with the components

dx j
v (x) =
j
.
dt
• The corresponding differential operator
n
d X ∂f
v( f )(x) = f (ϕt (x)) = v j (x) j
dt t=0
j=1
∂x

is the derivative along the streamline of the flow through the point p.

diffgeom.tex; April 12, 2006; 17:59; p. 25

22 CHAPTER 1. MANIFOLDS

• Conversely, every vector field v determines a flow, which is determined by

requiring v to be the velocity field of the flow.

• Such a flow is defined as the solution of the system of ordinary first-order

differential equations

dx j
= v j (x1 (t), . . . , xn (t)) , j = 1, . . . , n,
dt
with the initial conditions
x j (0) = x0j .
The solution of this system defines the integral curves of the vector field v.

Theorem 1.4.1 Fundamental Theorem on Vector Fields in Rn Let ⊂ Rn

be an open set in Rn and v : U → Rn be a smooth vector field on U.
Then for any point x0 ∈ U there is ε > 0 and a neighborhood V of x0
such that:

1. there is a unique curve (called the integral curve of v) x(t) =

φt (x0 ), t ∈ (−ε, ε) such that for any t ∈ (−ε, ε)
dx(t)
ẋ(t) ≡ = v(x(t))
dt
• and x(0) = x0 ;

2. the map
V × (−ε, ε) → Rn
defined by (x, t) 7→ ϕt (x) is smooth and for any t1 , t2 ∈ (−ε, ε)
such that t1 + t2 ∈ (−ε, ε)

ϕt1 ◦ ϕt2 = ϕt1 +t2

holds in V. The family of maps ϕt is called a local one-parameter

group of diffeomorphisms or a local flow.

• Remarks.

• The local flow is only defined in a small neighborhood of the point x0 .

• The one-parameter group is not a group in the strict sense.

diffgeom.tex; April 12, 2006; 17:59; p. 26

1.4. VECTOR FIELDS AND FLOWS 23

• The integral curves exist only for a small time.

• If the vector field is not differentiable, then the integral curve is not unique.

1.4.2 Vector Fields on Manifolds

• Let W be an open subset of a mnifold M and v be a smooth vector field on
W.

• Let (Uα , xα ) be a local chart in W.

• If W ⊂ Uα , then one can proceed as in Rn .

• If W is not contained in a single chart, then we choose a cover of W and

proceed as follows. Let p ∈ W and (Uα , xα ) and (Uβ , xβ ) be two charts
covering p.

• Then the integral curves in both local coordinate systems have the same
meaning and define a unique integral curve in M. This defines a local flow
on W in M. We just need to check that if the flow equations are satisfied in
one coordinate system, then they are satisfied in another coordinate system.

1.4.3 Straightening Flows

• Let U be an open set in a manifold M and ϕt : U → M be a local flow on
a M such that ϕ0 (p) = p. Then ϕt (p) depends smoothly on both the time t
and the point p.

• Note that if a vector field does not vanish at a point p, i.e. v p , 0, then it
does not vanish in a neighborhood of p.

• Let p0 be a fixed point in M and W be a sufficiently small hypersurface

passing through p0 such that v p is transversal to W at every point of p ∈ W.
This just means that v is nowhere tangent to W.

• If W is small enough it can be covered by a single coordinate chart. Let

(u1 , . . . , un−1 ) be local coordinates for W such that the the point p0 has coor-
dinates (0, . . . , 0).

• Then for a small neighborhood of a point p and sufficiently small t the n-

tuple (u1 , . . . , un−1 , t) gives a local coordinate system near p in M.

diffgeom.tex; April 12, 2006; 17:59; p. 27

24 CHAPTER 1. MANIFOLDS

• Proof: (follows from Inverse Function Theorem).

• Consider a map F : W × (−ε, ε) → M given by F(p, t) = ϕt (p).

• The differential of this map at p0 (that is, at u = 0), is

∂ ∂
F∗ = j, j = 1, . . . , n − 1 .
∂u j ∂u
Also,
d
F∗ v = ϕt (p0 ) = v .
dt
Thus
F∗ = Id ,
and, hence, (u1 , . . . , un−1 , t) can be used as a coordinate system near p0 .

• In these coordinates the flow is

ϕ s (u, t) = (u, t + s)

and
∂
v= .
∂t
• Thus, near a non-singular point of a vector field v one can introduce local
coordinates (u1 , . . . , un−1 , un ) such that

du j
= δnj
dt
that is,

du j dun
= 0, if j = 1, . . . , n − 1, and = 1.
dt dt

• Near a non-singular point of a vector field all flows are qualitatively the
same.

• Let v be a vector field on a manifold M. A point p ∈ M is called a singular

point of a vector field v if v p = 0 and a non-singular if v p , 0.

diffgeom.tex; April 12, 2006; 17:59; p. 28

Chapter 2

Tensors

2.1 Covectors and Riemannian Metric

2.1.1 Linear Functionals and Dual Space
• Let E be a real n-dimensional vector space.

• Let B = {e1 , . . . , en } be a basis in E.

• Then for any v ∈ E

n
X
v= v je j
j=1

• The real numbers (v1 , . . . , vn ) are called the components of v with respect
to the basis B.

• Remark. Every real vector space of dimension n is isomorphic to Rn .

Definition 2.1.1 Let E be a real vector space. A linear functional on

E is a linear real-valued function on E. That is, it is a map

α:E→R
•
satisfying the linearity conditions: ∀a, b, ∈ R and ∀v, w ∈ E

α(av + bw) = aα(v) + bα(w) .

25
26 CHAPTER 2. TENSORS

• Given a basis {e j }, j = 1, . . . , n, we define

a j = α(e j ) .
Then for any v ∈ E we have
n
X
α(v) = a jv j .
j=1

Definition 2.1.2 Let E be a real vector space. The set of all linear
•
functionals on E is called the dual space and denoted by E ∗ .

• The dual space is a real vector space under the natural operations of addition
and multiplication by scalar defined by: ∀α, β ∈ E ∗ , c1 , c2 ∈ R, v ∈ E,
(c1 α + c2 β)(v) = c1 α(v) + c2 β(v) .

• Let {e j } be a basis in E and {σ j }, j = 1, . . . , n, be linear functionals on E

such that
σ j (ei ) = δij .

• Then
σ j (v) = v j .

Theorem 2.1.1 Let E be a real vector space and {e j }, j = 1, . . . , n, be

a basis in E. Let and {σ j }, j = 1, . . . , n, be linear functionals on E such
that
σ j (ei ) = δij .
Then the linear functionals {σ j } for a basis in the dual space E ∗ , called
the dual basis to the basis {e j }, so that for any α ∈ E ∗
• n
X
α= a jσ j .
j=1

The real numbers

a j = α(e j )
are called the components of the linear functional α with respect to the
basis {σ j }.
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 29

2.1. COVECTORS AND RIEMANNIAN METRIC 27

1.


2.1.2 Differential of a Function

Definition 2.1.3 Let M be a manifold and p ∈ M be a point in M. The
• space T p∗ M dual to the tangent space T p M at p is called the cotangent
space.

Definition 2.1.4 Let M be a manifold and f : M → R be a real valued

smooth function on M. Let p ∈ M be a point in M. The differential
d f ∈ T p∗ M of f at p is the linear functional
•
d f : Tp → R

defined by
d f (v) = v p ( f ) .

• In local coordinates x j the differential is defined by

n
X ∂f
d f (v) = v j (x)
j=1
∂x j

In particular,
∂ ∂f
!
df = .
∂x j ∂x j
Thus
∂
!
dx i
= δij .
∂x j

and
dxi (v) = vi .

• The differentials {dxi } form a basis for the cotangent space T p∗ M called the
coordinate basis.
• Therefore, every linear functional has the form
n
X
α= a j dx j .
j=1

diffgeom.tex; April 12, 2006; 17:59; p. 30

28 CHAPTER 2. TENSORS

• That is why, the linear functionals are also called differential forms, or
1-forms, or covectors, or covariant vectors.
Definition 2.1.5 A covector field α is a differentiable assignment of
a covector α p ∈ T p∗ M to each point p of a manifold. This means that
•
the components of the covector field are differentiable functions of local
coordinates.
• Therefore, a covector field has the form
n
X
α= a j (x)dx j
j=1

• Under a change of local coordinates xαj = xαj (xβ ) the differentials transform
according to
n
X ∂xαj i
dxα =
j
dx .
i=1
∂xβi β

• Therefore, the components of a covector transform as

n
X ∂x j α β
aαi = a .
j=1
∂x i j
β

2.1.3 Inner Product

• Let E be a n-dimensional real vector space.
• The inner product (or scalar product) on E is a symmetric bilinear non-
degenerate functional on E × E, that is, it is a map h , i : E × E → R such
that
1. ∀v, w ∈ E,
hv, wi = hw, vi
2. ∀v, w, u ∈ E, ∀a, b ∈ R
hav + bu, wi = ahv, wi + bhu, wi

3. ∀v ∈ E,
if hv, wi = 0, ∀w ∈ E, then v = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 31

2.1. COVECTORS AND RIEMANNIAN METRIC 29

4. If, in addition, ∀v ∈ E,
hv, vi ≥ 0
and
hv, vi = 0 if and only if v = 0 ,
then the inner product is called positive definite.

• For a positive-definite inner product the norm of a vector v is defined by

p
||v|| = hv, vi

• Let {e j } be a basis in E.

• Then the matrix gi j defined by

gi j = hei , e j i

is a metric tensor, more precisely it gives the components of the metric

tensor in that basis.

• The matrix gi j is symmetric and nondegenerate, that is,

gi j = g ji , det gi j , 0 .

For a positive definite inner product, this matrix is positive-definite, that

is, it has only positive real eigenvalues. One says, that the metric has the
signature (+ · · · +). In special relativity one considers metrics wich are not
positive definite but have the signature (− + · · · +).

• Two vectors v, w ∈ E are orthogonal if

hv, wi = 0 .

• A vector u ∈ E is called unit vector if

||u|| = 1 .

• The basis is called orthonormal if

gi j = δi j .

diffgeom.tex; April 12, 2006; 17:59; p. 32

30 CHAPTER 2. TENSORS

• The inner product is given then by

n
X
hv, wi = gi j vi w j .
i, j=1

• Let v ∈ E. Then we can define a linear functional ν ∈ E ∗ by

ν(w) = hv, wi .

• Therefore, each vector v ∈ E defines a covector ν ∈ E ∗ called the covariant

version of the vector v.

• Given a basis {e j } in E and the dual basis σ j in E ∗ we find

n
X
νi = gi j v j .
j=1

• This operation is called lowering the index of a vector.

• Therefore, we can denote the components of the covector ν corresponding

to a vector v by the same symbol and call them the covariant components
of the vector.

• In an orthonormal basis, of course,

vi = vi .

• Similarly, given a covector ν ∈ E ∗ we can define a vector v ∈ E such that

∀w ∈ E ∗
ν(w) = hv, wi .

• Let gi j represent the matrix inverse to the matrix gi j .

• Then the contravariant components can be computed from the covariant

components by
Xn
v =
i
gi j v j .
j=1

This operation is called raising the index of a covector.

diffgeom.tex; April 12, 2006; 17:59; p. 33

2.1. COVECTORS AND RIEMANNIAN METRIC 31

• Thus, the vector spaces E and E ∗ are isomorphic. The isomorphism is pro-
vided by the inner product (the metric).

• The space E can also be considered as the space of linear functionals on E ∗ .

A vector v ∈ E defines a linear functional v : E ∗ → R by, for any α ∈ E ∗

v(α) = α(v).

2.1.4 Riemannian Manifolds

Definition 2.1.6 Let M be a manifold. A Riemannian metric on M is
a differentiable assignment of a positive definite inner product in each
tangent space T p M to the manifold at each point p ∈ M.

• If the inner product is non-degenerate but not positive definite, then it is

a pseudo-Riemannian metric.

A Riemannian manifold is a pair (M, g) of a manifold with a Rieman-

nian metric on it.

• Let p ∈ M be a point in M and (U, xα ) be a local coordinate system about p.

Let ∂i be the coordinate basis in T p M and gαi j = h∂αi , ∂αj i be the components
of the metric tensor in the coordinate system xα . Let (Uβ , xβ ) be another
coordinate system containing p. Then the components of the metric tensor
transform as
n ∂xk xl
β β β
X
α
gi j = g .
x j kl
i
k,l=1 α xα

Definition 2.1.7 Let (M, g) be a Riemannian manifold and f : M → R

be a smooth function. Then the gradient of f is a vector field grad f
• associated with the covector field d f . That is, for any p ∈ M and any
v ∈ TpM
h grad f, vi = d f (v) .

• In local coordinates,
n
X ∂f
( grad f ) =
i
gi j .
j=1
∂x j

diffgeom.tex; April 12, 2006; 17:59; p. 34

32 CHAPTER 2. TENSORS

2.1.5 Curves of Steepest Ascent

• Let (M, g) be a Riemannian manifold.
• Let p ∈ M be a point in M. For positive-definite inner product there is the
Schwartz inequality: for any v, w ∈ T p M,
|hv, wi| ≤ ||v|| ||w|| .

• Let u ∈ T p M be a unit vector. Then for any f ∈ C ∞ (M)

u( f ) = h grad f, ui .

• Therefore,
|u( f )| = |h grad f, ui| ≤ || grad f || .
• Thus, f has a maximum rate of change in the direction of the gradient.

Definition 2.1.8 Let a ∈ R be a real number. Then the level set of

f : M → R is the subset of M defined by
•
S = f −1 (a) = {p ∈ M | f (p) = a} .

• Let f : M → R be a smooth function, a ∈ R and S = f −1 (a) be the level set

of f . Let p0 ∈ S be a point in the level set S and let p = p(t), t ∈ (−ε, ε), be
a curve in S so that p(0) = p0 , f (p(t)) = a and ṗ(0) ∈ T p0 S . Then
d
f (p(t)) = h grad f, ṗi = d f ( ṗ) = 0 .
dt
• Thus, the gradient of f is orthogonal to the level set of f .
• The flow
dp grad f
=
dt || grad f ||
defined by the gradient of a smooth function f is called the Morse defor-
mation. It has the property that
df
=1
dt
and, therefore, in time t it maps the level set f −1 (a) into the level set f −1 (a +
t).

diffgeom.tex; April 12, 2006; 17:59; p. 35

2.2. TANGENT BUNDLE 33

2.2 Tangent Bundle

2.2.1 Fiber Bundles
• Let M and E be smooth manifolds and π : E → M be a smooth map. Then
the triple (E, π, M) is called a bundle.
• The manifold M is called the base manifold of the bundle and the manifold
E is called the bundle space of the bundle (or the bundle space manifold).
The map π is called the projection.
• The inverse image π−1 (p) of a point p ∈ M is called the fiber over p.
• The projection map is supposed to be surjective, that is, the differential π∗
has the maximal rank equal to dim M.
• Let {Uα }α∈A be an atlas of local charts covering the base manifold M and let
Uαβ = Uα ∩ Uβ , Uαβγ = Uα ∩ Uβ ∩ Uγ etc.
• A fiber bundle is a bundle all fibers of which, π−1 (p), ∀p ∈ M, are diffeo-
morphic to a common manifold F called the typical fiber of the bundle (or
just the fiber).
• For a fiber bundle, the inverse images π−1 (Uα ) are diffeomorphic to Uα × F.
That is, there are diffeomorhisms
hα : Uα × F → π−1 (Uα ) ,
such that for any p ∈ Uα ⊂ M, σ ∈ F
π(hα (p, σ)) = p .

The diffeomorphisms
ϕαβ = h−1
β ◦ hα : U αβ × F → U αβ × F

are called the transition functions of the bundle.

• The transition functions are defined by, ∀p ∈ Uαβ ⊂ M, σ ∈ F,
ϕαβ (p, σ) = (p, (Rαβ (p))(σ)) .
That is, for all p ∈ Uαβ there are diffeomorphisms Rαβ (p) of the fiber
Rαβ (p) : F → F .

diffgeom.tex; April 12, 2006; 17:59; p. 36

34 CHAPTER 2. TENSORS

• It is required that the set of all transformations Rαβ (p) ∈ G for all α, β and
p ∈ Uαβ ⊂ M forms a group G. This group is called the structure group of
the bundle.

• Thus, the transition functions ϕαβ determine smooth maps

Rαβ : Uαβ → G ,

that assign to each point p ∈ Uαβ an element Rαβ (p) ∈ G of the structure
group.

• Of course, from the definition of these maps we immediately obtain the

consistency conditions (or compatibility conditions)

Rαβ (p) = (Rβα (p))−1 , ∀p ∈ Uαβ ,

Rαβ (p)Rβγ (p)Rγα (p) = Id M , ∀p ∈ Uαβγ .

• A principal bundle is a fiber bundle (E, π, M) whose fiber F coincides with

the structure group, that is, F = G.

• A fiber bundle with any fiber F is fully determined by the transition func-
tions satisfying the consistency conditions. The fiber F does not play much
role in this construction.

• Let F be a manifold and Diff(F) be the set of all diffeomorphisms F → F.

Let G be a group and e ∈ G be the identity element of G. Then a map
T : G → Diff(F) such that

T (e) = IdF ,
T (R−1 ) = (T (R))−1 , ∀R ∈ G ,
T (R1 R2 ) = T (R1 ) ◦ T (R2 ) , ∀R1 , R2 ∈ G ,

is called a representation of the group G.

• Given a fiber bundle (E, π, M) with a fiber F and a structure group G one
can construct another fiber bundle (E 0 , π0 , M) with a fiber F 0 and the same
structure group G as follows. One takes a representation of the structure
group T : G → Diff(F 0 ) on the fiber F 0 and simply replaces the transition
functions Rαβ by T (Rαβ ). Such a fiber bundle is called a bundle associated
with the original bundle.

diffgeom.tex; April 12, 2006; 17:59; p. 37

2.2. TANGENT BUNDLE 35

• Thus, every fiber bundle is an associated bundle with some principal bun-
dle. So, all bundles can be constructed as associated bundles from principal
bundles. All we need is the structure group. The fiber is not important.

• A vector bundle is a fiber bundle whose fiber is a vector space.

• A section of a bundle (E, π, M) is a map s : M → E such that the image of

each point p ∈ M is in the fiber π−1 (p) over this point, that is, s(p) ∈ π−1 (p),
or
π ◦ s = Id M .

2.2.2 Tangent Bundle

Definition 2.2.1 Let M be a smooth manifold. The tangent bundle
T M to M is the collection of all tangent vectors at all points of M.
•
T M = {(p, v) | p ∈ M, v ∈ T p M}

• Let dim M = n.

• Let p ∈ M be a point in the manifold M, (U, x) be a local chart and (xi ) be

the local coordinates of the point p.

• Let ∂i = ∂/∂xi be the coordinate basis for T p M.

• Let v = ni=1 vi ∂i ∈ T p M. Then the local coordinates of the point (p, v) ∈

P
T M are
(x1 , . . . , xn , v1 , . . . , vn ) .

• Remarks.

• The coordinates (xi ) are local; they are restricted to the local chart U, that
is, (xi ) ∈ U ⊂ Rn .

• The coordinates vi are not restricted, that is, (vi ) ∈ Rn , they take any values
in Rn .

• The open set U × Rn ⊂ R2n is a local chart in the tangent bundle T M.

• Let (Uα , xα ) and (Uα , xα ) be two local charts containing the point p.

diffgeom.tex; April 12, 2006; 17:59; p. 38

36 CHAPTER 2. TENSORS

• Then the local coordinates of the point (p, v) in overlapping local charts are
related by

xαi = xαi (xβ )

n
X ∂xαi j
vα =
i
v
j β
j=1 ∂xβ

• This is a local diffeomorphism.

• Thus, the tangent bundle T M is a manifold of dimension 2 × dim M.

• A map π : T M → M defined by

π(p, v) = p

is called the projection map. It assigns to a vector tangent to M the point

in M at which the vector sits.

• Locally, if p has coordinates (x1 , . . . , xn ) and v has components (v1 , . . . , vn )

in the coordinate basis, then

π(x1 , . . . , xn , v1 , . . . , vn ) = (x1 , . . . , xn ) .

• Let p ∈ M be a point in M. The set π−1 (p) ⊂ T M is called the fiber of the
tangent bundle.

• The fiber of the tangent bundle

π−1 (p) = T p M

is the tangent space at p.

• Remarks.

• There is no global projection map π0 : T M → Rn defined by π0 (p, v) = v.

• In general, T M , M × Rn .

• For any chart U ∈ M

π−1 (U) = U × Rn .
Thus, locally the tangent bundle is a product manifold.

diffgeom.tex; April 12, 2006; 17:59; p. 39

2.2. TANGENT BUNDLE 37

• A vector field is a map

v : M → TM,
such that
π ◦ v = Id : M → M .

• A vector field is a cross section of the tangent bundle.

• The image of the manifold under a vector field is a n-dimensional subman-

ifold of the tangent bundle T M.

• The zero vector field defines the zero section of the tangent bundle.

Definition 2.2.2 Let (M, g) be an n-dimensional Riemannian mani-

fold. The unite tangent bundle of M is the set T 0 M of all unit vectors
• to M,
T 0 M = {(p, v) | p ∈ M, v ∈ T p M, ||v|| = 1} ,
where, locally, ||v||2 = ni, j=1 gi j (p)vi v j .
P

• The unit tangent bundle is a (2n−1)-dimensional submanifold of the tangent

bundle T M.

Theorem 2.2.1 Let S 2 be the unit 2-sphere embedded in R3 . The unit

tangent bundle T 0 S 2 is homeomorphic to the real projective space RP3
• and to the special orthogonal group S O(3)

T 0 S 2 ∼ RP3 ∼ S O(3) .
Proof :

1.


diffgeom.tex; April 12, 2006; 17:59; p. 40

38 CHAPTER 2. TENSORS

2.3 The Cotangent Bundle

Definition 2.3.1 Let M be a smooth manifold. The cotangent bundle
T ∗ M to M is the collection of all covectors at all points of M
•
T ∗ M = {(p, σ) | p ∈ M, σ ∈ T p∗ M}

• Let dim M = n.
• Let p ∈ M be a point in the manifold M, (U, x) be a local chart and (xi ) be
the local coordinates of the point p.
• Let dxi be the coordinate basis for T p∗ M.
• Let α = ni=1 αi dxi ∈ T p∗ M. Then the local coordinates of the point (p, α) ∈
P
T ∗ M are
(x1 , . . . , xn , α1 , . . . , αn ) .
• Remarks.
• The open set U × Rn ⊂ R2n is a local chart in the cotangent bundle T ∗ M.
• Let (Uα , xα ) and (Uα , xα ) be two local charts containing the point p.
• Then the local coordinates of the point (p, σ) in overlapping local charts are
related by
xαi = xαi (xβ )
n ∂x j
β β
X
α
σi = σj .
j=1
∂x i
α

• This is a local diffeomorphism.

• Thus, the cotangent bundle T ∗ M is a manifold of dimension 2n.
• The projection map π : T ∗ M → M is defined by π(p, α) = p.
• A covector field (or a 1-form)is a map
α : M → T∗M ,
such that
π ◦ α = Id : M → M .

diffgeom.tex; April 12, 2006; 17:59; p. 41

2.3. THE COTANGENT BUNDLE 39

• A covector field is a section of the cotangent bundle.

2.3.1 Pull-Back of a Covector

• Let M and N be two smooth manifolds and n = dim M and m = dim N.
• Let ϕ : M → N be a smooth map.
• The differential
ϕ∗ : T p M → T ϕ(p) N
is the linear transformation of the tangent spaces.
• Let xi be a local coordinate system in a local chart about p ∈ M and yα be a
local coordinate system in a local chart about ϕ(p) ∈ N and ∂i and ∂α be the
coordinate bases for T p M and T ϕ(p) N.
• Then the action of the differential ϕ∗ is defined by
m
∂ ∂yα ∂
! X
ϕ∗ = .
∂x j α=1
∂x j ∂yα

• Let v = vi ∂i . Then
Pn
i=1
n
X ∂yα
[ϕ∗ (v)]α = vj .
j=1
∂x j

Definition 2.3.2 The pullback ϕ∗ is the linear transformation of the

cotangent spaces
ϕ∗ : T ϕ(p)
∗
N → T p∗ M
taking covectors at ϕ(p) ∈ N to covectors at p ∈ M, defined as follows.
If α ∈ T ϕ(p)
∗
N, then ϕ∗ (α) ∈ T p∗ M so that
•
ϕ∗ (α) = α ◦ ϕ∗ : T p M → R

where α : T ϕ(p) N → R. That is, for any vector v ∈ T p M

(ϕ∗ (α)) (v) = α(ϕ∗ (v)) .

• Diagram.

diffgeom.tex; April 12, 2006; 17:59; p. 42

40 CHAPTER 2. TENSORS

• In local coordinates,
m
α
X ∂yα
[ϕ (dy )] j =
∗
dx j .
α=1
∂x j

• Let σ = σα dyα . Then

Pm
α=1
m
n X
X ∂yα j
ϕ (σ) =
∗
σα dx ,
j=1 α=1
∂x j

that is, in components,

m
X ∂yα
[ϕ (σ)] j =
∗
σα .
α=1
∂x j

• Remark.
• In general, for a map ϕ : M → N, the following linear transformations
are well defined: the differential ϕ∗ : T p M → T ϕ(p) N and the pullback
ϕ∗ : T ϕ(p)
∗
N → T p∗ M.
• The maps T p∗ M → T ϕ(p)
∗
N and T ϕ(p) N → T p M are not well defined, in
general.
• If dim M = dim N and ϕ : M → N is a diffeomorphism, then all these maps
are well defined.
• Explain.

2.3.2 Phase Space

• Let M be a configuration space of a dynamical system with local general-
i
ized coordinates q1 , . . . , qn . Then q̇i = dq
dt
are the generalized velocities.
• Under a change of local coordinates
qiα = qiα (qβ )
the velocities transform as components of a vector
n
X ∂q jα i
q̇αj = q̇
i β
.
i=1
∂qβ

diffgeom.tex; April 12, 2006; 17:59; p. 43

2.3. THE COTANGENT BUNDLE 41

• Therefore, (q1 , . . . , qn , q̇1 , . . . , q̇n ) give local coordinates for the tangent bun-
dle T M.

• Let L : T M → R be a map. Then L(q, q̇) is called the Lagrangian.

• The generalized momenta pi are defined by

∂L
pi = .
∂q̇i

• The momenta are functions on T M, that is, p : T M → R.

• Under a change of local coordinates qα = qα (qβ ) the momenta transform as

components of a covector
n ∂qi
β
pβi ,
X
pαj =
i=1 ∂qα
j

• The matrix
∂2 L
Hik (q, q̇) =
∂q̇i ∂q̇k
is called the Hessian.

• Suppose that the Hessian is non-degerate

det Hik , 0 .

• Then the velocities can be expressed in terms of momenta

q̇i = q̇i (q, p) ,

that is, there is a map q̇ : T ∗ M → R. More generally, there is a map

T M → T ∗ M.

• Thus, (q1 , . . . , qn , p1 , . . . , pn ) give local coordinates for the cotangent bundle

T ∗ M.

• The cotangent bundle is called the phase space in dynamics.

diffgeom.tex; April 12, 2006; 17:59; p. 44

42 CHAPTER 2. TENSORS

• The Hamiltonian is a smooth function on the cotangent bundle H : T ∗ M →

R defined by
n
X ∂L i
H(q, p) = q̇ − L(q, q̇) .
i=1
∂qi

• Example.

• One of the most important examples is the Lagrangian quadratic in veloci-

ties
n
1X
L(q, q̇) = gi j (q)q̇i q̇ j − V(q) ,
2 i, j=1

where gi j is a Riemannian metric on M and V is a smooth function on M.

• Then the Hessian is

∂2 L
gik =
∂q̇i ∂q̇k
and, therefore, nondegenerate.

• The relation between momenta and the velocities is

n
X n
X
pi = gi j (q)q̇ j , q̇i = gi j (q)p j .
j=1 j=1

• The Hamiltonian is given by

n
1 X ij
H(q, p) = g (q)pi p j + V(q) ,
2 i, j=1

2.3.3 The Poincaré 1-Form

• The Poincaré 1-form λ is a 1-form on the cotangent bundle T ∗ M defined
in local coordinates (q, p) on T ∗ M by
n
X
λ= pi dqi .
i=1

• Remarks.

diffgeom.tex; April 12, 2006; 17:59; p. 45

2.3. THE COTANGENT BUNDLE 43

• The coordinates p are not functions on M.

• The Poincaré form is not a 1-form on M.
• A general 1-form on T ∗ M is
Xn n
X
α= αi (q, p)dq +
i
vi (q, p)d pi .
i=1 i=1

Theorem 2.3.1 The Poincaré 1-form is well defined globally on the

•
cotangent bundle of any manifold.
Proof :
1. Let (qα , pα ) and (qβ , pβ ) be two overlapping coordinate patches of
T ∗ M.
2. Then n
X ∂qiα j
dqiα = dqβ
j=1 ∂q j
β
and n n
pβj dqβj .
X X
pαi dqiα =
i=1 j=1

• We give now an intrinsic definition of the Poincaré form.
• Let (q, p) ∈ T ∗ M be a point in T ∗ M. We want to define a 1-form λ ∈
∗
T (q,p) (T ∗ M) at this point (q, p) ∈ T ∗ M.
• Let π : T ∗ M → M be the projection defined for any q ∈ M, p ∈ T q∗ M by
π(q, p) = q.
• Then the pullback is the map π∗ : T q∗ M → T (q,p)
∗
(T ∗ M). For each 1-form
p ∈ T q∗ M it defines a 1-form π∗ (p) ∈ T (q,p)
∗
(T ∗ M). This is precisely the
Poincaré 1-form λ, that is,
λ = π∗ (p) .
• Of course, in local coordinates
n
X
π (p) =
∗
pi dqi .
i=1

diffgeom.tex; April 12, 2006; 17:59; p. 46

44 CHAPTER 2. TENSORS

2.4 Tensors
2.4.1 Covariant Tensors
• Let E be a vector space and E ∗ be its dual space.

• Let ei be a basis for E and σi be the dual basis for E ∗ .

• A covariant tensor of rank p (or a tensor of type (0, p)) is a multi-linear

real-valued functional
Q:| E × {z
··· × E
}→R
p

• Remarks.

• The function Q(v1 , . . . , v p ) is linear in each argument.

• The functional Q is independent of any basis.

• A covariant vector (covector) is a covariant tensor of rank 1.

• A metric tensor is a covariant tensor of rank 2.

• The components of the tensor Q with respect to the basis ei are defined
by
Qi1 ...i p = Q(ei1 , . . . , ei p ) .

• Then for any vectors

n
X
va = vaj e j ,
j=1

where a = 1, . . . , p, we have
n
X j
Q(v1 , . . . , v p ) = Q j1 ... j p v1j1 · · · v pp .
j1 ,..., j p =1

• The collection of all covariant tensors of rank p forms a vector space de-
noted by
Tp = | · · · ⊗ E}∗ .
E ∗ ⊗ {z
p

diffgeom.tex; April 12, 2006; 17:59; p. 47

2.4. TENSORS 45

• The dimension of the vector space T p is

dim T p = n p .

• The tensor product of two covectors α, β ∈ E ∗ is a covariant tensor α ⊗ β ∈

E ∗ ⊗ E ∗ of rank 2 defined by: ∀v, w ∈ E

(α ⊗ β)(v, w) = α(v)β(w) .

• The components of the tensor product α ⊗ β are

(α ⊗ β)i j = αi β j .

• The tensor product of a covariant tensor Q of rank p and a covariant ten-

sor T of rank q is a covariant tensor Q ⊗ T of rank (p + q) defined by:
∀v1 , . . . , v p , w1 , . . . , wq ∈ E

(Q ⊗ T )(v1 , . . . , v p , w1 , . . . , wq ) = Q(v1 , . . . , v p )T (w1 , . . . , wq ) .

• The components of the tensor product Q ⊗ T are

(Q ⊗ T )i1 ...i p j1 ... jq = Qi1 ...i p T j1 ... jq .

• Thus,
⊗ : T p × T q → T p+q .

• Tensor product is associative.

• The basis in the space T p is

σi1 ⊗ · · · ⊗ σi p ,

where 1 ≤ i1 , . . . , i p ≤ n.

• A covariant tensor Q of rank p has the form

n
X
Q= Qi1 ...i p σi1 ⊗ · · · ⊗ σi p .
i1 ,...,i p =1

diffgeom.tex; April 12, 2006; 17:59; p. 48

46 CHAPTER 2. TENSORS

2.4.2 Contravariant Tensors

• A contravariant vector can be considered as a linear real-valued functional

v : E∗ → R .

• A contravariant tensor of rank p (or a tensor of type (p, 0)) is a multi-

linear real-valued functional

E ∗ × {z
T :| · · · × E}∗ → R
p

• Remarks.

• The function T (α1 , · · · , α p ) is linear in each argument.

• The functional T is independent of any basis.

• A contravariant vector (covector) is a contravariant tensor of rank 1.

• The components of the tensor T with respect to the basis σi are defined
by
T i1 ...i p = T (σi1 , . . . , σi p ) .

• Then for any covectors

n
X
α(a) = α(a) j σ j ,
j=1

where a = 1, . . . , p, we have
n
X
T (α(1) , . . . , α(p) ) = T j1 ... j p α(1) j1 · · · α(p) j p .
j1 ,..., j p =1

• The inverse matrix of the components of a metric tensor defines a con-

travariant tensor g−1 of rank 2 by
n
X
g (α, β) =
−1
gi j αi β j .
i, j=1

diffgeom.tex; April 12, 2006; 17:59; p. 49

2.4. TENSORS 47

• The collection of all contravariant tensors of rank p forms a vector space

denoted by
Tp = |E ⊗ {z }.
··· ⊗ E
p

• The dimension of the vector space T p is

dim T p = n p .

• The tensor product of a contravariant tensor Q of rank p and a contravari-

ant tensor T of rank q is a contravariant tensor Q ⊗ T of rank (p + q) defined
by: ∀α1 , . . . , α p , β1 , . . . , βq ∈ E ∗

(Q ⊗ T )(α1 , . . . , α p , β1 , . . . , βq ) = Q(α1 , . . . , α p )T (β1 , . . . , βq ) .

• The components of the tensor product Q ⊗ T are

(Q ⊗ T )i1 ...i p j1 ... jq = Qi1 ...i p T j1 ... jq .

• Thus,
⊗ : T p × T q → T p+q .

• Tensor product is associative.

• The basis in the space T p is

ei1 ⊗ · · · ⊗ ei p ,

where 1 ≤ i1 , . . . , i p ≤ n.

• A contravariant tensor T of rank p has the form

n
X
T= T i1 ...i p ei1 ⊗ · · · ⊗ ei p .
i1 ,...,i p =1

• The set of all tensors of type (p, 0) forms a vector space T p of dimension n p

dim T p = n p .

diffgeom.tex; April 12, 2006; 17:59; p. 50

48 CHAPTER 2. TENSORS

2.4.3 General Tensors of Type (p, q)

• A tensor of type (p, q) is a multi-linear real-valued functional

E ∗ × {z
T :| · · · × E}∗ × |
E × {z
··· × E
}→R
p q

• The components of the tensor T with respect to the basis ei , σi are defined
by
i ...i
T j11 ... jpq = T (σi1 , . . . , σi p , e j1 , . . . , e jq ) .

• Then for any covectors

n
X
α(a) = α(a) j σ j ,
j=1

where a = 1, . . . , p, and any vectors

n
X
vb = vkb ek ,
j=1

where b = 1, . . . , q, we have
n n
j ... j
X X
T (α(1) , . . . , α(p) , v1 , . . . , vq ) = T k11...kqp α(1) j1 · · · α(p) j p vk1 · · · vkq .
k1 ,...,kq =1 j1 ,..., j p =1

• The inverse matrix of the components of a metric tensor defines a con-

travariant tensor g−1 of rank 2 by
n
X
g−1 (α, β) = gi j αi β j .
i, j=1

• The collection of all tensors of type (p, q) forms a vector space denoted by

T qp = |
E ⊗ {z
··· ⊗ E
}⊗| · · · ⊗ E}∗ .
E ∗ ⊗ {z
p q

• The dimension of the vector space T p is

dim T qp = n p+q .

diffgeom.tex; April 12, 2006; 17:59; p. 51

2.4. TENSORS 49

• The tensor product of a tensor Q of type (p, q) and a tensor T of type (r, s)
is a tensor Q⊗T of type (p+r, q+s) defined by: ∀α1 , . . . , α p , β1 , . . . , βr ∈ E ∗ ,
v1 , . . . , vq , w1 , . . . , w s ∈ E

(Q ⊗ T )(α1 , . . . , α p , β1 , . . . , βr , v1 , . . . , vq , w1 , . . . , w s )
= Q(α1 , . . . , α p , v1 , . . . , vq )T (β1 , . . . , βr , w1 , . . . , w s ) .

• The components of the tensor product Q ⊗ T are

i ...i j ... j i ...i

... jr
(Q ⊗ T )k11 ...kpq l11 ...lrs = Qk11 ...kpq T lj11...l s
.

• Thus,
p+r
⊗ : T qp × T sr → T q+s .

• We stress once again that the tensor product is associative.

• The basis in the space T qp is

ei1 ⊗ · · · ⊗ ei p ⊗ σ j1 ⊗ · · · ⊗ σ jq ,

where 1 ≤ i1 , . . . , i p , j1 , . . . , jq ≤ n.

• A tensor T of type (p, q) has the form

n n
i ...i
X X
T= T j11 ... jpq ei1 ⊗ · · · ⊗ ei p ⊗ σ j1 ⊗ · · · ⊗ σ jq .
j1 ,..., jq =1 i1 ,...,i p =1

• Let p, q ≥ 1 and 1 ≤ r ≤ p, 1 ≤ s ≤ q. The (r, s)-contraction of tensors of

type (p, q) is the map
p−1
tr rs : T qp → T q−1

defined by
n
i ...i i ...i ki ...i
X
(tr rs T ) j11 ... jp−1
q−1
= T j11 ... jr−1 r p−1
s−1 k j s ... jq
.
k=1

diffgeom.tex; April 12, 2006; 17:59; p. 52

50 CHAPTER 2. TENSORS

2.4.4 Linear Transformations and Tensors

• Let A : E → E be a linear transformation.
• Then we can define a tensor à of type (1, 1) by ∀α ∈ E ∗ , v ∈ E
Ã(α, v) = α(Av) .

• Let (Ai j ) be the matrix of the linear transformation A, that is,

n
X
Ae j = Ai j ei .
i=1

• Then the components of the tensor à are

Ãi j = Ã(σi , e j ) = σi (Ae j ) = Ai j .

• Thus, one can identify tensors of type (1, 1) and linear transformations on
E (and, similarly on E ∗ as well).
• Then,
n
X
A= Ai j ei ⊗ σ j .
i, j=1

• The identity linear transformation I has the matrix

I i j = δij
and defines the tensor of type (1, 1)
n
X
I= ei ⊗ σi .
i=1

• To summarize, a (1, 1) tensor à : E ∗ × E → R is identified with the linear

transformation A : E → E.
• The covariant tensor Ai j , the contravariant tensor Ai j and the tensor Ai j of
type (1, 1) are related by
n
X n
X
Ai j = gik Ak j , Ai j = Ai k gk j .
k=1 k=1

diffgeom.tex; April 12, 2006; 17:59; p. 53

2.4. TENSORS 51

2.4.5 Tensor Fields

Definition 2.4.1 A tensor field on a manifold M is a smooth assign-
•
ment of a tensor at each point of M.

• Let xαi = xαi (xβ ) be a local diffeomorphism.

• Then
n
X ∂xi α
dxαi = dxβj
j=1 ∂xβ
j

and
∂ X ∂xβ ∂n j

=
∂xαi j=1
∂xαi ∂xβj

• Let T be a tensor of type (p, q). Then

∂xαp ∂xβ ∂xβj1

n n i l1
i ...i
X X ∂xαi1 k ...k
T (α) j11 ... jpq = ··· ··· T (β) l11...lqp
k1 ,...,k p =1 l1 ,...,lq =1 ∂xβ
k1 k l j
∂xβp ∂xαq ∂xαq

2.4.6 Tensor Bundles

Definition 2.4.2 Let M be a smooth manifold. The tensor bundle of
type (p, q) T qp M is the collection of all tensors of type (p, q) at all points
• of M
T qp M = {(p, T ) | p ∈ M, T ∈ T q,p (x) M}

• The tensor bundle T qp M is the tensor product of the tangent and cotangent
bundles
T qp M = T
| M ⊗ {z
··· ⊗ TM}⊗T
∗
| M ⊗ {z }.
· · · ⊗ T∗M
p q

• Let dim M = n.

• Let x ∈ M be a point in the manifold M, (U, x) be a local chart and (xi ) be

the local coordinates of the point p.

• Let ∂i and dxi be the coordinate basis for T x M and T x∗ M.

diffgeom.tex; April 12, 2006; 17:59; p. 54

52 CHAPTER 2. TENSORS

i ...i
• Let T be a tensor of type (p, q) and T j11 ... jpq be the components of the tensor T
in the coordinate basis. Then the local coordinates of the point (p, T ) ∈ T qp M
are
i ...i
(xi , T j11 ... jpq ) ,
where 1 ≤ i, i1 , . . . , i p , j1 , . . . , jq ≤ n.

• Remarks.
p+q p+q
• The open set U × Rn ⊂ Rn+n is a local chart in the tensor bundle T qp M.

• The bundle T qp M is a manifold of dimension n + n p+q .

• The projection map π : T qp M → M is defined by π(x, T ) = x.

• A tensor field of type (p, q) is a map

T : M → T qp M ,

such that
π ◦ T = Id : M → M .

• A tensor field of type (p, q) is a section of the tensor bundle T qp M.

2.4.7 Examples
• Riemannian metric gµν .

• Energy-momentum tensor T µν .

• Stress tensor σi j .

• Riemann curvature tensor Rµ αβγ .

• Ricci curvature tensor Rµν .

• Scalar density of weight 1: |g| =

p
det gi j .

• Axial vectors (vector product) in R3 .

• Strength of the electromagnetic field Fµν .

diffgeom.tex; April 12, 2006; 17:59; p. 55

2.4. TENSORS 53

Theorem 2.4.1 Let n

X
A= Ai dxi
i=1

be a covector field (1-form). Let

• Fi j = ∂i A j − ∂ j Ai .

Then n
X
F= Fi j dxi ⊗ dx j
i, j=1

is a tensor of type (0, 2).

Proof :

1. Check the transformation law.



• A tensor is called isotropic if it is a tensor product of g, g−1 and I.

• The components of an isotropic tensor are the products of gi j , gi j and δij .

• Every isotropic tensor of type (p, q) has an even rank p + q.

• For example, the most general isotropic tensor of type (2, 2) has the form

Ai j kl = agi j gkl + bδik δlj + cδil δkj ,

where a, b, c are scalars.

2.4.8 Einstein Summation Convention

• In any expression there are two types of indices: free indices and repeated
indices.

• Free indices appear only once in an expression; they are assumed to take all
possible values from 1 to n.

• The position of all free indices in all terms in an equation must be the same.

diffgeom.tex; April 12, 2006; 17:59; p. 56

54 CHAPTER 2. TENSORS

• Repeated indices appear twice in an expression. It is assumed that there is a

summation over each repeated pair of indices from 1 to n. The summation
over a pair of repeated indices in an expression is called the contraction.

• Repeated indices are dummy indices: they can be replaced by any other
letter (not already used in the expression) without changing the meaning of
the expression.

• Indices cannot be repeated on the same level. That is, in a pair of repeated
indices one index is in upper position and another is in the lower position.

• There cannot be indices occuring three or more times in any expression.

diffgeom.tex; April 12, 2006; 17:59; p. 57

Chapter 3

Differential Forms

3.1 Exterior Algebra

3.1.1 Permutation Group
• A group is a set G with an associative binary operation, · : G × G → G with
identity, called the multiplication, such that each element has an inverse.
That is, the following conditions are satisfied

1. for any three elements g, h, k ∈ G, the associativity law holds: (gh)k =

g(hk);
2. there exists an identity element e ∈ G such that for any g ∈ G, ge =
eg = g;
3. each element g ∈ G has an inverse g−1 , such that g g−1 = g−1 g = e

• Let X be a set. A transformation of the set X is a bijective map g : X → X.

• The set of all transformations of a set X forms a group Aut(X), with com-
position of maps as group multiplication.

• Any subgroup of Aut(X) is a transformation group of the set X.

• The transformations of a finite set X are called permutations.

• The group S p of permutations of the set Zn = {1, . . . , p} is called the sym-

metric group of order p.

55
56 CHAPTER 3. DIFFERENTIAL FORMS

• Theorem 3.1.1 The order of the symmetric group S p is

|S p | = p! .

• Any subgroup of S p is called a permutation group.

• A permutation ϕ : Z p → Z p can be represented by
1 ...
!
p
ϕ(1) . . . ϕ(p)

• The identity permutation is

1 ...
!
p
1 ... p

• The inverse permutation ϕ−1 : Z p → Z p is represented by

ϕ(1) . . . ϕ(p)
!
1 ... p

• The product of permutations is then defined in an obvious manner.

• An elementary permutation is a permutation that exchanges the order of
only two elements.
• Every permutation can be realized as a product of elementary permutations.
• A permutation that can be realized by an even number of elementary per-
mutations is called an even permutation.
• A permutation that can be realized by an odd number of elementary permu-
tations is called an odd permutation.
• Proposition 3.1.1 The parity of a permutation does not depend on the
representation of a permutation by a product of the elementary ones.
• That is, each representation of an even permutation has even number of
elementary permutations, and similarly for odd permutations.
• The sign of a permutation ϕ, denoted by sign(ϕ) (or simply (−1)ϕ ), is
defined by
+1, if ϕ is even,
(
ϕ
sign(ϕ) = (−1) =
−1, if ϕ is odd

diffgeom.tex; April 12, 2006; 17:59; p. 58

3.1. EXTERIOR ALGEBRA 57

3.1.2 Permutations of Tensors

• Let S p be the symmetric group of order p. Then every permutation ϕ ∈ S p
defines a map
ϕ : Tp → Tp ,
which assigns to every tensor T of type (0, p) a new tensor ϕ(T ), called a
permutation of the tensor T , of type (0, p) by: ∀v1 , . . . , v p
ϕ(T )(v1 , . . . , v p ) = T (vϕ(1) , . . . , vϕ(p) ) .

• Let (i1 , . . . , i p ) be a p-tuple of integers. Then a permutation ϕ : Z p → Z p

defines an action
ϕ(i1 , . . . , i p ) = (iϕ(1) , . . . , iϕ(p) ) .

• The components of the tensor ϕ(T ) are obtained by the action of the permu-
tation ϕ on the indices of the tensor T
ϕ(T )i1 ...i p = T iϕ(1) ...iϕ(p) .

• The symmetrization of the tensor T of the type (0, p) is defined by

1 X
Sym(T ) = ϕ(T ) .
p! ϕ∈S
p

• The symmetrization is also denoted by parenthesis. The components of the

symmetrized tensor Sym(T ) are given by
1 X
T (i1 ...i p ) = T i ...i .
p! ϕ∈S ϕ(1) ϕ(p)
p

• The anti-symmetrization of the tensor T of the type (0, p) is defined by

1 X
Alt(T ) = sign (ϕ)ϕ(T ) .
p! ϕ∈S
p

• The anti-symmetrization is also denoted by square brackets. The compo-

nents of the anti-symmetrized tensor Alt(T ) are given by
1 X
T [i1 ...i p ] = sign (ϕ)T iϕ(1) ...iϕ(p) .
p! ϕ∈S
p

diffgeom.tex; April 12, 2006; 17:59; p. 59

58 CHAPTER 3. DIFFERENTIAL FORMS

• Examples.

• A tensor T of type (0, p) is called symmetric if for any permutation ϕ ∈ S p

ϕ(T ) = T .

• A tensor T of type (0, p) is called anti-symmetric if for any permutation

ϕ ∈ Sp
ϕ(T ) = sign (ϕ)T .

• An anti-symmetric tensor of type (0, p) is called a p-form.

• Remarks.

• Permutation, symmetrization, anti-symmetrization of tensors of type (p, 0).

• Completely symmetric and completely anti-symmetric tensors of type (p, 0).

• An anti-symmetric tensor of type (p, 0) is called a p-vector.

• Partial permutation.

• Examples.

• Notation.

3.1.3 Alternating Tensors

• Let (i1 , . . . , i p ) and ( j1 , . . . , j p ) be two p-tuples of integers 1 ≤ i1 , . . . , i p , j1 , . . . , j p ≤
n. The generalized Kronecker symbol is defined by

1 if (i1 , . . . , i p ) is an even permutation of ( j1 , . . . , j p )



i ... i

δ j11 ... jpp =  if (i1 , . . . , i p ) is an odd permutation of ( j1 , . . . , j p )

−1


 0 otherwise


• One can easily check that

δij11 . . . δij1p
 
 
i ... i .. . . .
δ j11 ... jpp = det  . ..
 
. 

i i
δ jp1 . . . δ jpp


diffgeom.tex; April 12, 2006; 17:59; p. 60

3.1. EXTERIOR ALGEBRA 59

• Also, there holds

i ... i i
δ j11 ... jpp = p!δi[1j1 · · · δ jpp ] .
i ... i
• Thus, the Kronecker symbols δ j11 ... jpp are the components of the tensors
p!Alt(I|⊗ {z
· · · ⊗}I )
p

of type (p, p), which are anti-symmetric separately in upper indices and the
lower indices.
• Thus, the anti-symmetrization can also be written as
1 j1 ... j p
T [i1 ...i p ] = δ T j ... j .
p! i1 ...i p 1 p

• Notation.
• Obviously, the Kronecker symbols vanish for p > n
i ...i
δ j11 ... jpp = 0 if p > n .

• The contraction of Kronecker symbols gives Kronecker symbols with lower

indices, more precisely, we have the theorem.

Theorem 3.1.2 For any p, q ∈ N, 1 ≤ p, q ≤ n, there holds

• i ... i l ...l (n − p)! i1 ... i p
δ j11 ... jpp l11 ...lqq = δ .
(n − q)! j1 ... j p
Proof :
1.

Corollary 3.1.1
For any q ∈ N, 1 ≤ q ≤ n we have

i ...i n!
• δi11 ...iqq = .
(n − q)!
In particular,
δii11 ...i
...in = n! .
n

diffgeom.tex; April 12, 2006; 17:59; p. 61

60 CHAPTER 3. DIFFERENTIAL FORMS

Lemma 3.1.1 There holds

• i ...i j ... j i ...i k ...k
δl11 ...l pp m11 ...mrr δkj11 ......kjrr = r!δl11 ...l pp m11 ...mrr
Proof :
1.

• In general, let A be a p-form (an antisymmetric tensor of type (0, p)) and B
be a p-vector (an anti-symmetric tensor of type (p, 0)). Then
i ...i
Ai1 ...i p δ j11 ... jpp = p!A j1 ... j p ,
j ... j
Bi1 ...i p δi11 ... ipp = p!B j1 ... j p .

• Let (i1 , . . . , in ) be an n-tuple of integers 1 ≤ i1 , . . . , in ≤ n. The completely

anti-symmetric (alternating) Levi-Civita symbols are defined by

εi1 ...in = δi11......inn , εi1 ...in = δi11......inn ,

so that
if (i1 , . . . , in ) is an even permutation of (1, . . . , n)


 1
i1 ...in
ε = εi1 ...in = if (i1 , . . . , in ) is an odd permutation of (1, . . . , n)

−1


 0 otherwise


• Theorem 3.1.3 There holds the identity

X
εi1 ...in ε j1 ... jn = sign(ϕ) δij1ϕ(1) · · · δijnϕ(n)
ϕ∈S n

= n!δi[1j1 · · · δijnn ]
= δij11...i
... jn .
n

The contraction of this identity over k indices gives

εi1 ...in−k m1 ...mk ε j1 ... jn−k m1 ...mk = k!(n − k)!δi[1j1 · · · δijn−k

n−k ]

= k!δij11...i
... jn−k .
n−k

In particular,
εm1 ...mn εm1 ...mn = n! .

diffgeom.tex; April 12, 2006; 17:59; p. 62

3.1. EXTERIOR ALGEBRA 61

• It is easy to see that there holds also

i ...i
δl11 ...ln−p
n−p
ε j1 ... j p l1 ...ln−p = (n − p)!ε j1 ... j p i1 ...in−p

• The set of all n × n real matrices is denoted by Mat(n, R).

• The determinant is a map det : Mat(n, R) → R that assigns to each matrix

A = (Ai j ) a real number det A defined by
X
det A = sign (ϕ)A1 ϕ(1) · · · An ϕ(n) ,
ϕ∈S n

• The most important properties of the determinant are listed below:

Theorem 3.1.4 1. The determinant of the product of matrices is equal to

the product of the determinants:

det(AB) = det A det B .

2. The determinants of a matrix A and of its transpose AT are equal:

det A = det AT .

3. The determinant of the inverse A−1 of an invertible matrix A is equal

to the inverse of the determinant of A:

det A−1 = (det A)−1

4. A matrix is invertible if and only if its determinant is non-zero.

• The determinant of a matrix A = (Ai j ) can be written as

det A = εi1 ...in A1 i1 . . . An in

= ε j1 ... jn A j1 1 . . . A jn n
1 i1 ...in
= ε ε j1 ... jn A j1 i1 . . . A jn in .
n!
Here, as usual, a summation over all repeated indices is assumed from 1 to
n.

diffgeom.tex; April 12, 2006; 17:59; p. 63

62 CHAPTER 3. DIFFERENTIAL FORMS

3.1.4 Exterior p-forms

• An exterior p-form (or simply a p-form) is an anti-symmetric covariant
tensor α ∈ T p of type (0, p).

• The collection of all p-forms forms a vector space Λ p , which is a vector

subspace of T p
Λp ⊂ T p .

• In particular,
Λ0 = R and Λ1 = T 1 = E ∗ .

• In other words, a 0-form is a smooth function, and a 1-form is a covector

field.

• Let α ∈ Λ p be a p-form.

• Let ei be a basis in E and σi be the dual basis in E ∗ .

• The components of the p-form α are

αi1 ...i p = α(ei1 , . . . ei p ) .

• The components are completely anti-symmetric in all indices, that is,

αi1 ...i p = sign (ϕ)αiϕ(1) ...iϕ(p) .

In particular, under a permutation of any two indices the form changes sign

α... i ... j ... = −α... j ... i ... ,

which means that the components vanish if any two indices are equal

α... i ... i ... = 0 (no summation!) .

• Thus, all non-vanishing components have different indices.

• Therefore, the values of all components αi1 ...i p are completely determined by
the values of the components with the indices i1 , . . . , i p reordered in strictly
increasing order
1 ≤ i1 < · · · < i p ≤ n .

diffgeom.tex; April 12, 2006; 17:59; p. 64

3.1. EXTERIOR ALGEBRA 63

• Notation.
• To deal with forms it is convenient to introduce multi-indices. We will
denote a p-tuple of integers from 1 to n by a capital letter
I = (i1 , . . . , i p ) ,
where 1 ≤ i1 , . . . , i p ≤ n. For a p-tuple of the same integers ordered in an
increasing order we define
Iˆ = (i1 , . . . , i p ) .
where 1 ≤ i1 < i2 < · · · < i p ≤ n. We call Iˆ an increasing p-tuple associated
with I.
• Therefore, a collection of p-forms
 
p!Alt σi1 ⊗ · · · ⊗ σi p ,
where 1 ≤ i1 < i2 < · · · < i p ≤ n, forms a basis in the space Λ p .
• Thus, every p-form α ∈ Λ p has the form
X  
α= αi1 ···i p p!Alt σi1 ⊗ · · · ⊗ σi p .
1≤i1 <···<i p ≤n

• Therefore, the dimension of the space Λ p is equal to the number of distinct

increasing p-tuples of integers from 1 to n.

Theorem 3.1.5 The dimension of the space Λ p of p-forms is

• n
!
n!
dim Λ p = = .
p p!(n − p)!
Proof :
1.

• In particular,
dim Λ0 = dim Λn = 1 ,
dim Λ1 = dim Λn−1 = n ,
etc.

diffgeom.tex; April 12, 2006; 17:59; p. 65

64 CHAPTER 3. DIFFERENTIAL FORMS

• There are no p-forms with p > n.

• Similarly to the norm of vectors and covectors we define the inner product
of exterior p-forms α and β in a Riemannian manifold by
1 i1 j1
(α, β) = g · · · gi1 j p αi1 ...i p β j1 ... j p .
p!
• This enables one to define also the norm of an exterior p-form α by
||α|| = (α, α)
p

3.1.5 Exterior Product

• Since the tensor product of two skew-symmetric tensors is not a skew-
symmetric tensor to define the algebra of antisymmetric tensors we need to
define the anti-symmetric tensor product called the exterior (or wedge)
product.
• If α is an p-form and β is an q-form then the exterior product of α and β is
an (p + q)-form α ∧ β defined by
(p + q)!
α∧β= Alt (α ⊗ β) .
p!q!
• In components
(p + q)!
(α ∧ β)i1 ...i p+q = α[i1 ...i p βi p+1 ...i p+q ] .
p!q!
• Let α ∈ Λ p be a p-form. Then
p = deg(α)
is called the degree (or rank) of α.
Theorem 3.1.6 The exterior product has the following properties

(α ∧ β) ∧ γ = α ∧ (β ∧ γ) (associativity)
•
α ∧ β = (−1)deg(α)deg(β) β ∧ α (anticommutativity)

(α + β) ∧ γ = α ∧ γ + β ∧ γ (distributivity) .
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 66

3.1. EXTERIOR ALGEBRA 65

1.

• The exterior square of any p-form α of odd degree p (in particular, for any
1-form) vanishes
α ∧ α = 0.
• The exterior algebra Λ (or Grassmann algebra) is the set of all forms of
all degrees, that is,
Λ = Λ0 ⊕ · · · ⊕ Λn .
• The dimension of the exterior algebra is
n !
X n
dim Λ = = 2n .
p=0
p

• A basis of the space Λ p is

σ i1 ∧ · · · ∧ σ i p , (1 ≤ i1 < · · · < i p ≤ n) .
An p-form α can be represented in one of the following ways
α = αi1 ...i p σi1 ⊗ · · · ⊗ σi p
1
= αi1 ...i p σi1 ∧ · · · ∧ σi p
p!
X
= αi1 ...i p σi1 ∧ · · · ∧ σi p .
i1 <···<i p

• The exterior product of a p-form α and a q-form β can be represented as

1
α∧β= α[i1 ...i p βi p+1 ...i p+q ] σi1 ∧ · · · ∧ σi p+q .
p!q!

Theorem 3.1.7 Let σ j ∈ Λ1 , 1 ≤ j ≤ n, and α j ∈ Λ1 , 1 ≤ j ≤ n, be

two collections of n 1-forms related by a linear transformation
n
X
• αj = A j i σi , 1 ≤ j ≤ n,
i=1

Then
α1 ∧ · · · ∧ αn = det Ai j σ1 ∧ · · · ∧ σn .
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 67

66 CHAPTER 3. DIFFERENTIAL FORMS

1.

Theorem 3.1.8 Let α j ∈ Λ1 = E ∗ , 1 ≤ j ≤ p, be a collections of p
1-forms and vi ∈ E, 1 ≤ i ≤ p, be a collection of p vectors. Let

• A j i = α j (vi ) , 1 ≤ i, j ≤ p .

Then  
α1 ∧ · · · ∧ α p (v1 , . . . , v p ) = det Ai j .
Proof :
1.

Theorem 3.1.9 A collections of p 1-forms α j ∈ Λ1 = E ∗ , 1 ≤ j ≤ p,
• is linearly dependent if and only if

α1 ∧ · · · ∧ α p = 0 .

Corollary 3.1.2 Let xi = xi (x0 ), i = 1, . . . , n, be a local diffeomor-

phism. Then
•
∂xl
!
dx ∧ · · · ∧ dx = det
1 n
dx01 ∧ · · · ∧ dx0n .
∂x 0m

3.1.6 Interior Product

• The interior product of a vector v and a p-form α is a (p − 1)-form iv α
defined by, for any v1 , . . . , v p−1 ,

iv α(v1 , . . . , v p−1 ) = α(v, v1 , . . . , v p−1 ) .

In particular, if p = 1, then iv α is a scalar

iv α = α(v)

and if p = 0, then by definition

iv α = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 68

3.1. EXTERIOR ALGEBRA 67

• In components,
(iv α)i1 ...i p−1 = v j α ji1 ...i p−1 .

• The interior product is a map

iv : Λ p → Λ p−1 ,

or
iv : Λ → Λ .

• A map L : Λ → Λ is called an derivation if for any α ∈ Λ p , β ∈ Λq ,

L(α ∧ β) = (Lα) ∧ β + α ∧ Lβ .

• A map L : Λ → Λ is called an anti-derivation if for any α ∈ Λ p , β ∈ Λq ,

L(α ∧ β) = (Lα) ∧ β + (−1) p α ∧ Lβ .

Theorem 3.1.10 Let v ∈ E be a vector. The interior product iv : Λ →

•
Λ is an anti-derivation.

diffgeom.tex; April 12, 2006; 17:59; p. 69

68 CHAPTER 3. DIFFERENTIAL FORMS

3.2 Orientation and the Volume Form

3.2.1 Orientation of a Vector Space
• Let E be a vector space. Let {ei } = {e1 , . . . , en } and {e0j } = {e01 , . . . , e0n } be two
different bases in E related by

ei = Λ j i e0j ,

where Λ = (Λi j ) is a transformation matrix.

• Note that the transformation matrix is non-degenerate

det Λ , 0 .

• Since the transformation matrix Λ is invertible, then the determinant det Λ

is either positive or negative.

• If det Λ > 0 then we say that the bases {ei } and {e0i } have the same orien-
tation, and if det Λ < 0 then we say that the bases {ei } and {e0i } have the
opposite orientation.

• If the basis {ei } is continuously deformed into the basis {e0j }, then both bases
have the same orientation.

• Since det I = 1 > 0 and the function det : GL(n, R) → R is continuous, then
a one-parameter continuous transformation matrix Λ(t) such that Λ(0) = I
preserves the orientation.

• This defines an equivalence relation on the set of all bases on E called the
orientation of the vector space E.

• This equivalence relation divides the set of all bases in two equivalence
classes, called the positively oriented and negatively oriented bases.

• A vector space together with a choice of what equivalence class is positively

oriented is called an oriented vector space.

diffgeom.tex; April 12, 2006; 17:59; p. 70

3.2. ORIENTATION AND THE VOLUME FORM 69

3.2.2 Orientation of a Manifold

• Let M be a manifold and let T p M be the tangent space at a point p ∈ M.

• Let (U, x) be a local coordinate patch about a point p ∈ M.

• Then the vectors

∂
, i = 1, . . . , n
∂xi
form a basis in T p M.

• Let (U 0 , x0 ) be another local coordinate system about a point p, that is, there
is a local diffeomorphism xi = xi (x0 ).

• Then the vectors

∂ ∂x j ∂
=
∂x0i ∂x0i ∂x j
form another basis in T p M.

• The orientation of the bases {∂i } and {∂0j } is the same (or consistent) if

∂xi
!
det > 0.
∂x0 j

• If it is possible to choose an orientation of all tangent spaces T p M at all

points in a continuous fashion, then the orientation of all tangent spaces is
consistent.

• A manifold M is called orientable if there is an atlas such that the ori-

entation of all charts of this atlas can be chosen consistently, that is, the
Jacobians of all transition functions have positive determinant.

• Each connected orientable manifold has exactly two possible orientations.

One orientation can be declared positive, then the other orientation is neg-
ative.

• An orientable manifold with a chosen orientation is called oriented.

• Remarks.

• If a manifold can be covered by a single coordinate chart then it is ori-

entable.

diffgeom.tex; April 12, 2006; 17:59; p. 71

70 CHAPTER 3. DIFFERENTIAL FORMS

• Not all manifolds are orientable.

• Transport of the orientation.

• Let p, q ∈ M be two points in a manifold M and C(t) be a curve in M

connecting p and q, i.e.

C(0) = p, C(1) = q .

• Let {ei (t)} be a basis in TC(t) M that continuously depends on t ∈ [0, 1].

• Then the orientation of the basis ei (1) is uniquely determined by the orien-
tation of the basis ei (0).

• Thus, the orientation is transported along a curve in a unique way.

• Note that the transportation of the basis is not unique, in general. Only the
transportation of the orientation is!

• Given a point p and another point q, the orientation at the point q does, in
general, depend on the curve C(t) connecting the points p and q.

• If a manifold is orientable, then the transportation of the orientation from

one point to another does not depend on the curve connecting the points.

Corollary 3.2.1 Let M be a manifold. If there exist two points p and

q in M and two curves C1 (t) and C2 (t) joining the points p and q such
that the orientation at q transported from p along the curves C1 and C2
• are different, then M is nonorientable.

That is, if there exists a closed curve C(t) in M such that the transport
of the orientation along C leads to a reversal of orientation, then M is
nonorientable.
• Example. Möbius Band.

3.2.3 Hypersurfaces in Orientable Manifolds

• Let V ⊂ W be a submanifold of a manifold W.

• A vector field N on W is transverse to V if N is nowhere tangent to V.

diffgeom.tex; April 12, 2006; 17:59; p. 72

3.2. ORIENTATION AND THE VOLUME FORM 71

• That is, N is transverse to V if for any point p ∈ V, N p < T p V.

• If N is transverse to V, then N , 0 on V.

• Let W be an n-diemsnional manifold. An (n − 1)-diemsnional submanifold

M of W is called a hypersurface in W.

• A hypersurface M in W is two-sided in W if there is a continuous vector N

in W transverse to M.

• Examples. Normals to surfaces in R3 .

Theorem 3.2.1 A two-sided hypersurface in an orientable manifold is

•
orientable.
• Remarks.

• Orientability of a manifold is an intrinsic property of a manifold.

• Two-sidedness of a manifold depends on the embedding of the manifold as

a hypersurface in a higher-dimensional manifold.

• Example.

• Every manifold (even a nonorientable one) M is a two-sided hypersurface

in a manifold M × R.

3.2.4 Projective Spaces

• The real projective space RP2 is the sphere S 2 with antipodal points identi-
fied.

• The sphere S 2 is two-sided in R3 , so it is orientable.

• Let e1 , e2 be a basis in T p S 2 and N be an outward pointing normal vector to

S 2.

• Then {N, e1 , e2 } is a basis in R3 .

• We say that the basis e1 , e2 is positively oriented in S 2 if the basis {N, e1 , e2 }

is positive in R3 .

diffgeom.tex; April 12, 2006; 17:59; p. 73

72 CHAPTER 3. DIFFERENTIAL FORMS

• Let
r : R3 → R3
be the reflection map defined by

r(x) = −x .

• Let
a : S2 → S2
be the antipodal map in S 2 defined as the restriction of the reflection map to
S2
a = r S2

• The reflection map reverses the orientation of the basis in R3 !

• Let p ∈ S 2 be a point on S 2 and {N, e1 , e2 } ∈ T p S 2 be a positively oriented

basis. Then {−e2 , −e1 , −N} is negatively oriented at T a (p)S 2 , where a(p) is
the antipodal point.

• The vector −N is the outward normal at the antipodal point a(p).

• Thus the basis {−e1 , −e2 } is negatively oriented at a(p).

• Thus, if the basis {e1 , e2 } at p ∈ S 2 is transported along a curve C on S 2 to

a(p), then the orientation of the transported basis {e01 , e02 } is the opposite of
the orientation of the basis {−e1 , −e2 } at a(p).

• Thus, the antipodal map reverses the orientation on S 2 .

• In RP2 the basis {−e1 , −e2 } at a(p) represents the same basis {e1 , e2 } at p.

• Thus, the curve C on S 2 is a closed curve C 0 on RP2 going through p.

• Thus, the transportation of the basis along the closed curve C 0 in RP2 re-
verses the orientation.

• Thus, RP2 is not orientable!

• More generally, the even-dimensional projective spaces RP2n are not ori-
entable.

• The odd-dimensional projective spaces RP2n are orientable.

diffgeom.tex; April 12, 2006; 17:59; p. 74

3.2. ORIENTATION AND THE VOLUME FORM 73

3.2.5 Pseudotensors and Tensor Densities

• Let E be an oriented vector space and B be the set of all bases on E. Then
the orientation is a function

o:B→R

defined by
 +1 if ei is positively oriented



o(ei ) = 


 −1 if ei is negatively oriented


• A pseudo-tensor T on a vector space E assigns, for each orientation o of

E a tensor T o such that when the orientation is reversed the tensor changes
sign, i.e.
T −o = −T o .

• That is, a pseudo-tensor is a collection of two tensors T + and T − , one for

each orientation.

• A pseudo-tensor field on a manifold is a smooth assignment of a pseudo-

tensor to each point of the manifold.

• Let xαi = xαi (xβ ) be a local diffeomorphism and

 
 ∂xi 
J(xβ ) = det  αj  .
∂xβ

• Since this transformation is a diffeomorphism J , 0. Thus, there are two

cases J > 0 and J < 0.

• A pseudo-tensor of type (p, q) is a geometric object T whose components

i ...i
T j11 ... jpq in the coordinate basis ∂i and dxi transform according to

∂xαp ∂xβ ∂xβj1

n n i l1
i ...i
X X ∂xαi1 k ...k
T (α) j11 ... jpq (xα ) = sign (J) ··· ··· T (β) l11...lqp (xβ )
k1 ,...,k p =1 l1 ,...,lq =1 ∂xβ
k1 k l j
∂xβp ∂xαq ∂xαq

diffgeom.tex; April 12, 2006; 17:59; p. 75

74 CHAPTER 3. DIFFERENTIAL FORMS

• A tensor density of weight w of type (p, q) is a geometric object T whose

i ...i
components T j11 ... jpq in the coordinate basis ∂i and dxi transform under a dif-
feomorphism xαi = xαi (xβ ) with J > 0 according to

∂xαp ∂xβ ∂xβj1

n n i l1
i ...i
X X ∂xαi1 k ...k
T (α) j11 ... jpq (xα ) = J (xβ )
−w
··· ··· T (β) l11...lqp (xβ )
k1 ,...,k p =1 l1 ,...,lq =1 ∂xβ
k1 k l j
∂xβp ∂xαq ∂xαq

Theorem 3.2.2 Let g = (gi j ) be a Riemannian metric and

• |g| = det(gi j ) .
p
Then |g| is a scalar density of weight 1.

Proof :

1. Let x = x(x0 ) be a local diffeomorphism.

2. Then
∂xk ∂xl
g0i j = gkl ,
∂x0i ∂x0 j

3. Then
!2
∂x
|g | = det
0
|g| .
∂x0

4. Thus
∂x p
!
|g | = det |g| .
p
0
∂x0

p
5. Therefore, |g| is a scalar density of weight 1.

• The Levi-civita symbol are not tensors!

diffgeom.tex; April 12, 2006; 17:59; p. 76

3.2. ORIENTATION AND THE VOLUME FORM 75

Theorem 3.2.3 Let gi j be the components of a Riemannian metric, |g| =

det(gi j ), and εi1 ...in and εi1 ...in be the Levi-Civita symbols and Ei1 ...in and
E i1 ...in be defined by
1
Ei1 ...in = |g| εi1 ...in , E i1 ...in = p εi1 ...in
p
|g|

Then
•
1. εi1 ...in represents the components of a pseudo-n-form (that is, a
pseudo-tensor density of type (0, n)) of weight (−1).

2. εi1 ...in represents the components of a pseudo-n-vector (that is, a

pseudo-tensor density of type (n, 0)) of weight 1.

3. Ei1 ...in represents the components of a pseudo-n-form.

4. E i1 ...in represents the components of a pseudo-n-vector.

Proof :
1. Check the transformation law.


3.2.6 Volume Form

• Let {v(1) , . . . , v(n) } be an ordered n-tuple of vectors. The volume of the par-
allelepiped spanned by the vectors {v(1) , . . . , v(n) } is a real number defined
by q
|vol (v(1) , . . . , v(n) )| = det((v(i) , v( j) )) .

• If the vectors {v(1) , . . . , v(n) } are orthonormal, then

|vol (v(1) , . . . , v(n) )| = 1 .

• If the vectors {v(1) , . . . , v(n) } are linearly dependent, then

vol (v(1) , . . . , v(n) ) = 0 .

• The volume |vol (v(1) , . . . , v(n) )| does not depend on the orientation of vec-
tors.

diffgeom.tex; April 12, 2006; 17:59; p. 77

76 CHAPTER 3. DIFFERENTIAL FORMS

• The signed volume of the parallelepiped spanned by an ordered n-tuple of

vectors {v(1) , . . . , v(n) } is

vol (v(1) , . . . , v(n) ) = sign(v(1) , . . . , v(n) ) |vol (v(1) , . . . , v(n) )| ,

where the sign of the signed volume is defined by

 +1, if {v(1) , . . . , v(n) } is positively oriented



sign(v(1) , . . . , v(n) ) = 

 −1, if {v(1) , . . . , v(n) } is negatively oriented


Theorem 3.2.4 Let ei be a basis in a vector space E, σi be the dual ba-

sis of 1-forms in E ∗ , g = (gi j ) be a Riemannian metric and {v(1) , . . . , v(n) }
be a set of n vectors. Let vi ( j) = σi (v( j) ) be the contravariant components
of the vectors v( j) and vi ( j) = (ei , v( j) ) = gik vk ( j) be the covariant compo-
nents of these vectors. Then:

vol (v(1) , . . . , v(n) ) =

p
|g| det(vi ( j) ))

= Ei1 ...in vi1 (1) · · · vin (n)

• and
1
vol (v(1) , . . . , v(n) ) = p det(vi ( j) )
|g|

= E i1 ...in vi1 (1) · · · vin (n)

In particular,
vol (e1 , . . . , en ) = o(ei ) |g| ,
p

where o(ei ) is equal to +1 or −1 for positively oriented and negatively

oriented basis ei .

• The n-form
vol = |g| σ1 ∧ · · · ∧ σn
p

is called the Riemannian volume element (or volume form).

• We have
i ...i
 
σi1 ∧ · · · ∧ σi p (e j1 , . . . , e j p ) = δ j11 ··· jpp .

diffgeom.tex; April 12, 2006; 17:59; p. 78

3.2. ORIENTATION AND THE VOLUME FORM 77

• The components of the volume form are

vol (ei1 , . . . , ein ) = |g|εi1 ...in = Ei1 ...in .

p

3.2.7 Star Operator and Duality

• The volume form allows one to define the duality of p-forms and (n − p)-
vectors.

• For each p-form Ai1 ...i p one assigns the dual (n − p)-vector by
1 j1 ... jn−p i1 ...i p
à j1 ... jn−p = E Ai1 ...i p .
p!

• Similarly, for each p-vector Ai1 ...i p one assigns the dual (n − p)-form by
1
à j1 ... jn−p = E j ... j i ...i Ai1 ...i p .
p! 1 n−p 1 p

• By lowering and raising the indices of the dual forms we can define the
duality of forms and poly-vectors separately.

• The Hodge star operator

∗ : Λ p → Λn−p

maps any p-form α to a (n − p)-form ∗α dual to α defined as follows.

• For each p-form α the form ∗α is the unique (n − p)-form such that

α ∧ ∗α = (α, α)vol .

• In particular,
∗1 = vol , ∗vol = 1 .

• In components, this means that

1
(∗α)i p+1 ... in = εi1 ... i p i p+1 ...in |g|gi1 j1 · · · gi p j p α j1 ... j p
p
p!
1 1
= p gi p+1 j p+1 · · · gin jn ε j1 ... j p j p+1 ... jn α j1 ... j p .
p! |g|

diffgeom.tex; April 12, 2006; 17:59; p. 79

78 CHAPTER 3. DIFFERENTIAL FORMS

Theorem 3.2.5 For any p-form α there holds

∗2 α = (−1) p(n−p) α .
•
In particular, if n is odd, then for any p

∗2 = Id .

Theorem 3.2.6 Let α be a 1-form and v be the corresponding vector,

• that is, vi = gi j α j . Then
∗α = iv vol .

• A collection {ω(1) , . . . , ω(n−1) } of (n − 1) 1-forms defines a 1-form α by

α = ∗ ω(1) ∧ · · · ∧ ω(n−1) .
 

• Then
∗α = (−1)n−1 ω(1) ∧ · · · ∧ ω(n−1)
and
ω(1) ∧ · · · ∧ ω(n−1) ∧ α = (α, α)vol .

• In components,
α j = g jk E i1 ...in−1 k ωi1 (1) · · · ωin−1 (n−1) .

• If the 1-forms {ω(1) , . . . , ω(n−1) } are linearly dependent, then α = 0.

• If the collection of 1-forms {ω(1) , . . . , ω(n−1) } is linearly independent, then

{ω(1) , . . . , ω(n−1) , α} are linearly independent and form a basis in E ∗ .

Theorem 3.2.7 Let {σ1 , . . . , σn } be an orthonormal basis of 1-forms.

Then
h i
• σ j = (−1)n− j ∗ σ1 ∧ · · · ∧ σ j−1 ∧ σ j+1 ∧ · · · ∧ σn ,

In particular,
σn = ∗ [σ1 ∧ · · · ∧ σn−1 ] .

• Similarly, a collection {v(1) , . . . , v(n−1) } of (n − 1) vectors defines a covector

α by: for any vector v

α(v) = vol (v(1) , . . . , v(n−1) , v)

diffgeom.tex; April 12, 2006; 17:59; p. 80

3.2. ORIENTATION AND THE VOLUME FORM 79

or, in components,

α j = Ei1 ...in−1 j vi1 (1) · · · vin−1 (n−1) .

Theorem 3.2.8 Let {v(1) , . . . , v(n−1) } be an ordered (n − 1)-tuple of lin-

early independent vectors and {ω(1) , . . . , ω(n−1) } be the corresponding
1-forms. Let α be 1-form defined by

α = ∗ ω(1) ∧ · · · ∧ ω(n−1) ,
 

and N be the corresponding vector, that is,

N i = gik |g|εi1 ...in−1 k vi1 (1) · · · vin−1 (n−1)

p

Then:
•
1. The vector N is orthogonal to all vectors {v(1) , . . . , v(n−1) }, that is,

(N, v( j) ) = gik N i vk ( j) = 0 , ( j = 1, . . . , n − 1) .

2. The n-tuple {v1 , . . . , vn−1 , N} forms a positively oriented basis.

3. The volume of the parallelepiped spanned by the vectors

{v(1) , . . . , v(n−1) , N} is determined by the norm of N

vol (v1 , . . . , vn−1 , N) = (N, N) .

diffgeom.tex; April 12, 2006; 17:59; p. 81

80 CHAPTER 3. DIFFERENTIAL FORMS

3.3 Exterior Derivative

• From now on, if not specified otherwise, we will denote the derivatives by
∂
∂i = .
∂xi
• The exterior derivative of a 0-form (that is, a function) f is a 1-form d f
defined by: for any vector v
(d f )(v) = v( f ) .

• In local cordinates
d f = ∂ j f dx j .

• The exterior derivative of a 1-form (that is, a covector) A is a 2-form dA

defined by: for any vectors v, w
(dA)(v, w) = v(A(w)) − w(A(v)) − A([v, w]) .

• In local cordinates
1 
dA = ∂i A j − ∂ j Ai dxi ∧ dx j .
2
• Let α be a p-form
1
α= αi1 ...i p dxi1 ∧ · · · ∧ dxi p .
p!

• The exterior derivative of α is a (p + 1)-form dα defined by

1
dα = dαi ...i dxi1 ∧ · · · ∧ dxi p
p! 1 p
1
= ∂i αi ...i dxi1 ∧ dxi2 ∧ · · · ∧ dxi p+1 .
p! 1 2 p+1

• In components
(dα)i1 i2 ...i p+1 = (p + 1)∂[i1 αi2 ...i p+1 ]
p+1
X
= (−1)k−1 ∂ik αi1 ...ik−1 ik+1 ...i p+1
k=1

diffgeom.tex; April 12, 2006; 17:59; p. 82

3.3. EXTERIOR DERIVATIVE 81

Theorem 3.3.1 The exterior derivative is a linear map

•
d : Λ p → Λ p+1 .
Proof : Show that dα is a form whose value does not depend on the coordi-
nate system.


Theorem 3.3.2 Let α ∈ Λ p be a p-form and {v1 , . . . , v p+1 } be a collec-

tion of (p + 1) vectors. Then
p+1
X
• (dα)(v1 , . . . , v p+1 ) = (−1)k−1 vk (α(v1 , . . . , vk−1 , vk+1 , . . . , v p+1 ))
k=1
p+1 X
X k−1
− (−1)i+k−1 α([vi , vk ], v1 , . . . , vi , vi+1 , . . . , vk−1 , vk+1 , . . . , v p+1 ))
k=1 i=1

Proof : Calculation.


• Remark. This formula can be taken as the intrinsic definition of the exterior
derivative.

Theorem 3.3.3 For any p-form

•
d2 = 0 .
Proof : Easy.


Theorem 3.3.4 The exterior derivative d : Λ → Λ is an anti-

derivation on the exterior algebra.
•
That is, for any p-form α ∈ Λ p and any q-form β ∈ Λq there holds

d(α ∧ β) = (dα) ∧ β + (−1) p α ∧ (dβ) .

Proof : Calculation.


• Examples.

diffgeom.tex; April 12, 2006; 17:59; p. 83

82 CHAPTER 3. DIFFERENTIAL FORMS

3.3.1 Coderivative
• Given a Riemannian metric gµν we also define the co-derivative of p-forms.
• The coderivative is a linear map

δ : Λ p → Λ p−1

defined by
δ = ∗−1 d∗ = (−1)(n−p+1)(p−1) ∗ d∗

• That is the coderivative of a p-form α is the (p − 1)-form δα defined by

1
(δα)i1 ...i p−1 = εi1 ...i p−1 i p ...in |g|g ji p g j p+1 i p+1 · · · g jn in
p
(n − p + 1)!
!
1
(n − p + 1)∂ j ε j ... j j ... j |g|g · · · g αk1 ...k p
p j1 k1 jpkp
p! 1 p p+1 n

Theorem 3.3.5 For any p-form

•
δ2 = 0 .
Proof : Follows from ∗2 = ±1 and d2 = 0.

• From this definition, we can also see that, for any 0-form f (a function) ∗ f
is an n-form and, therefore, d ∗ f = 0, i.e. a coderivative of any 0-form is
zero
δf = 0.

• For a 1-form α, δα is a 0-form

1 p 
δα = p ∂i |g|gi j α j .
|g|

• More generally, one can prove that for a p-form α

1 p 
(δα)i1 ...i p−1 = gi1 j1 . . . gi p−1 j p−1 p ∂ j |g|g jk g j1 k1 · · · g j p−1 k p−1 αkk1 ...k p−1 .
|g|

• Examples.

diffgeom.tex; April 12, 2006; 17:59; p. 84

3.4. PULLBACK OF FORMS 83

3.4 Pullback of Forms

• Let M be a n-dimensional manifold and W be a r-dimensional manifold.
• Let F : M → W be a smooth map of a manifold M to a manifold W.
• Let p ∈ M be a point in M and q = F(p) ∈ W be the image of p in W.
• Let xi , (i = 1, . . . , n), be a local coordinate system about p and yµ , (µ =
1, . . . , r), be a local coordinate system about q so that
yµ = yµ (x) .

• Let f : W → R be a smooth function on W.

• The pullback of f to M is a function F ∗ f : M → R on M defined by
F∗ f = f ◦ F ,
that is, for any x
(F ∗ f )(x) = f (y(x)) .
• Suppose that n = r and the map F is bijective.
• Let h : M → R be a smooth function on M.
• The pushforward of h to W is a function F∗ h : W → R on W defined by
F∗ h = h ◦ F −1 ,
that is, for any y
(F∗ h)(y) = h(x(y)) .
• Remarks. The pullback is well defined for an arbitrary map F.
• The pushforward is only defined for bijections!
• The pullback is the map
F ∗ : ΛpW → Λp M
defined as follows.
Let α ∈ Λ p W be a p-form on W. The pullback of α is a p-form F ∗ α on M
defined by: for any vectors v1 , . . . , v p
(F ∗ α)(v1 , . . . , v p ) = α(F∗ v1 , . . . , F∗ v p ) .

diffgeom.tex; April 12, 2006; 17:59; p. 85

84 CHAPTER 3. DIFFERENTIAL FORMS

• In local cordinates
1
F ∗α = αµ1 ...µ p (y(x))dyµ1 ∧ · · · ∧ dyµ p
p!
1 ∂yµ1 ∂yµ p
= · · · αµ1 ...µ p (y(x))dxi1 ∧ · · · ∧ dxi p
p! ∂x i 1 ∂x i p

• In components
∂yµ1 ∂yµ p
(F α)i1 ...i p
∗
= i1 · · · i p αµ1 ...µ p (y(x))
∂x ∂x

• Remark. The pullback is well-defined only for covariant tensors and the
pushforward is well defined only for contravariant tensors.

Theorem 3.4.1 Let F : M → W. The pullback F ∗ : Λ p W → Λ p M

has the properties:

1. F ∗ is linear.

2. For any two forms α and β

•
F ∗ (α ∧ β) = (F ∗ α) ∧ (F ∗ β)

3. F ∗ commutes with exterior derivative. That is, for any p-form α

F ∗ (dα) = d(F ∗ α) .
Proof : Direct calculation.


diffgeom.tex; April 12, 2006; 17:59; p. 86

3.5. VECTOR ANALYSIS IN R3 85

3.5 Vector Analysis in R3

3.5.1 Vector Algebra in R3
• In the case of three-dimensional Euclidean space the metric in Cartesian
coordinates is gi j = δi j .

• The bases of p-forms are:

1, dx, dy, dz, dx ∧ dy, dx ∧ dz, dy ∧ dz, dx ∧ dy ∧ dz .

• The star operator acts on this forms by

∗1 = dx ∧ dy ∧ dz,

∗dx = dy ∧ dz, ∗dy = −dx ∧ dz, ∗dz = dx ∧ dy,

∗(dx ∧ dy) = dz, ∗(dy ∧ dz) = dx, ∗(dx ∧ dz) = −dy,
∗(dx ∧ dy ∧ dz) = 1 .

• So, any 2-form

α = α12 dx ∧ dy + α13 dx ∧ dz + α23 dy ∧ dz

is represented by the dual 1-form

∗α = α12 dz − α13 dy + α23 dx ,

• That is
1
(∗α)i = εi jk α jk
2
(∗α)1 = α23 , (∗α)2 = α31 , (∗α)3 = α12 ,

• Any 3-form α
α = α123 dx ∧ dy ∧ dz
is represented by the dual 0-form
1
∗α = εi jk αi jk = α123 .
3!

diffgeom.tex; April 12, 2006; 17:59; p. 87

86 CHAPTER 3. DIFFERENTIAL FORMS

• Now, let α and β be two 1-forms

α = α1 dx + α2 dy + α3 dz , β = β1 dx + β2 dy + β3 dz .

Then
∗β = β1 dy ∧ dz + β2 dz ∧ dx + β3 dx ∧ dz
and

α ∧ β = (α1 β2 − α2 β1 )dx ∧ dy + (α1 β3 − α3 β1 )dx ∧ dz + (α2 β3 − α3 β2 )dy ∧ dz ,

α ∧ (∗β) = (α1 β1 + α2 β2 + α3 β3 )dx ∧ dy ∧ dz .

Therefore,

∗(α ∧ β) = (α1 β2 − α2 β1 )dz − (α1 β3 − α3 β1 )dy + (α2 β3 − α3 β2 )dx ,

∗[α ∧ (∗β)] = α1 β1 + α2 β2 + α3 β3 ,
or
∗(α ∧ β) = α × β ,
∗[α ∧ (∗β)] = α · β .

3.5.2 Vector Analysis in R3

• Zero-Forms.
For a 0-form f we have
(d f )i = ∂i f ,
so that
d f = grad f .

• One-Forms.
For a 1-form
α = α1 dx + α2 dy + α3 dz
we have
(dα)i j = ∂i α j − ∂ j αi
that is,

dα = (∂1 α2 − ∂2 α1 )dx ∧ dy + (∂2 α3 − ∂3 α2 )dy ∧ dz + (∂3 α1 − ∂1 α3 )dz ∧ dx .

diffgeom.tex; April 12, 2006; 17:59; p. 88

3.5. VECTOR ANALYSIS IN R3 87

• Therefore
(∗dα)i = εi jk ∂ j αk ,
so that

∗dα = (∂2 α3 − ∂3 α2 )dx + (∂3 α1 − ∂1 α3 )dy + (∂1 α2 − ∂2 α1 )dz .

• We see that
∗dα = curl α .

• Two-Forms.
For a 2-form β there holds

(dβ)i jk = ∂i β jk + ∂ j βki + ∂k βi j ,

or
dβ = (∂1 β23 + ∂2 β31 + ∂3 β12 )dx ∧ dy ∧ dz .

• Hence,
1
∗dβ = εi jk ∂i β jk = ∂1 β23 + ∂2 β31 + ∂3 β12 .
2
• Now let α be a 1-form

α = α1 dx + α2 dy + α3 dz .

Then
∗α = α1 dy ∧ dz − α2 dx ∧ dz + α3 dx ∧ dy ,
and
d ∗ α = (∂1 α1 + ∂2 α2 + ∂3 α3 )dx ∧ dy ∧ dz ,
or
∗d ∗ α = ∂1 α1 + ∂2 α2 + ∂3 α3 .
So,
∗d ∗ α = div α .

diffgeom.tex; April 12, 2006; 17:59; p. 89

88 CHAPTER 3. DIFFERENTIAL FORMS

diffgeom.tex; April 12, 2006; 17:59; p. 90

Chapter 4

Integration of Differential Forms

4.1 Integration over a Parametrized Subset

4.1.1 Integration of n-Forms in Rn
• Let U ⊂ Rn be a closed ball in Rn and ui , i = 1, . . . , n be the Cartesian
coordinates in Rn .
• Let f : U → R be a continuous real-valued function on U.
• Then the integral of f over U is the multiple integral
Z Z
f = f (u) du1 · · · dun .
U U

• Let o be an orientation of U so that o(u) = +1 if the coordinate basis of cov-

ectors (du1 , . . . , dun ) has the same orientation as o and o(u) = −1 otherwise.
• Let
α = f (u) du1 ∧ · · · ∧ dun
be an n-form.
• Then the integral of α over U is defined by
Z Z
α = o(u) f (u) du1 . . . dun .
U U

• The integral of the form α over U reverses sign if the orientation of U is

reversed.

89
90 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

4.1.2 Integration over Parametrized Subsets

• Let M be an n-dimensional manifold with local coordinates xi , i = 1, . . . , n.
• Let 0 ≤ p ≤ n and U be an oriented region in R p with orientation o and
coordinates uµ , µ = 1, . . . , p.
• Let F : U → M be a smooth map given locally by
xi = F i (u) .

• Then the image F(U) ⊂ M of the set U is called a p-subset of M and the
collection (U, o, F) is called an oriented parametrized p-subset of M.
• A parametrized 1-subset is called a curve in M.
• A parametrized 2-subset is called a surface in M.
• Remarks.
• A p-subset is not a submanifold, in general.
• A p-subset could have different dimensions at different points.
• Usually, the differential F∗ has rank p (that is, F(U) is a submanifold) al-
most everywhere.
• Let α ∈ Λ p be a p-form on M
1
α= αi ...i dxi1 ∧ · · · ∧ dxi p .
p! 1 p
• The integral of α over an oriented parametrized p-subset F(U) is defined
by Z Z
α= F ∗α .
F(U) U

• In more detail,
∂ ∂
Z Z !
α = o(u) (F α) ∗
, . . . , p du1 · · · du p
F(U) U ∂u1 ∂u
∂ ∂
Z !
= o(u) α F∗ 1 , . . . , F∗ p du1 · · · du p
U ∂u ∂u
∂x i1
∂xi p
Z
1
= o(u) αi1 ...i p (x(u)) 1 · · · p du1 · · · du p .
p! U ∂u ∂u

diffgeom.tex; April 12, 2006; 17:59; p. 91

4.1. INTEGRATION OVER A PARAMETRIZED SUBSET 91

4.1.3 Line Integrals

• Let U = [a, b] ⊂ R be an interval.

• Then a map F : U → M defines an oriented curve C = F(U) in M

xi = F i (t) .

• Let α be a 1-form in M
α = αi (x) dxi .

• Then the integral of α over C is called the line integral.

• In more detail
b b
dxi
Z Z " !# Z
d
α= α F∗ = αi (x(t)) dt .
C a dt a dt

4.1.4 Surface Integrals

• Let U ⊂ R2 be an oriented region in the plane, for example, U = [a, b] ×
[c, d].

• Then a map F : U → M defines an oriented parametrized surface S = F(U)

in M
xi = F i (u1 , u2 ) .

• Let α be a 2-form in M
1
α = αi j (x) dxi ∧ dx j .
2

• Then the integral of α over C is called the surface integral.

• In more detail
dxi dx j 1 2
Z Z " ! !# Z
d d 1
α= α F∗ , F ∗ du 1
du 2
= α ij (x(u)) du du .
S U du1 du2 U 2 du1 du2

• Example in R3 .

diffgeom.tex; April 12, 2006; 17:59; p. 92

92 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

4.1.5 Independence of Parametrization

• Let U, V ⊂ R p be regions in R p with coordinates uµ , µ = 1, . . . , p, and vν ,
ν = 1, . . . , p, respectively.

• Let H : U → V be an diffeomorphism so that V = H(U) defined by

vµ = H µ (u) .

• Then for any function f : V → R there holds a formula for the change of
variables in multiple integrals

∂(v1 , . . . , v p )
Z Z
f (v) dv . . . dv =
1 p
f [v(u)] du1 . . . du p
V U ∂(u1 , . . . , u p )

• Let M be an n-dimensional manifold with n ≥ p.

• Let U and V be oriented regions, uµ and vν be positively-oriented coordi-

nates on U and V, and the diffeomorphism H be orientation-preserving, that
is, the Jacobian
∂(v1 , . . . , v p )
>0
∂(u1 , . . . , u p )
is positive.

• Let F : U → M and G : V → M be a smooth maps so that

F =G◦H.

• Then F(U) is an oriented parametrized p-subset of M and G is a reparametriza-

tion of this subset
xi = F i (u) = Gi (H(u)) .

• Let α ∈ Λ p be a p-form on M

1
α= αi ...i (x) dxi1 ∧ · · · ∧ dxi p
p! 1 p

diffgeom.tex; April 12, 2006; 17:59; p. 93

4.1. INTEGRATION OVER A PARAMETRIZED SUBSET 93

• Then
Z Z
α = G∗ α
G(V) V
∂xi1 ∂xi p
Z
1
= αi1 ...i p (G(v)) 1 · · · p dv1 · · · dv p
p! V ∂v ∂v
∂x i1
∂xi p ∂(v1 , . . . , v p ) 1
Z
1
= αi1 ...i p (G(H(u))) 1 · · · p du · · · du p
p! U ∂v ∂v ∂(u1 , . . . , u p )
∂xi1 ∂xi p
Z
1
= αi1 ...i p (F(u)) 1 · · · p du1 · · · du p
p! ∂u ∂u
Z U
= α
U

• Thus, the integral is independent of the parametrization of a p-subset.

4.1.6 Integrals and Pullbacks

• Let M be an n-dimensional manifold and W be an r-dimensional manifold.

• Let ϕ : M → W be a smooth map.

• Let U ⊂ R p be an oriented region in Rn and F : U → M be an oriented

parametrized p-subset of M.

• Then ψ = ϕ ◦ F : U → W is an oriented parametrized p-subset of W.

• Let α ∈ Λ p W be a p-form on W.

• Then
Z Z Z Z Z
α= ψα=
∗
(F ◦ ϕ )α =
∗ ∗
F (ϕ α) =
∗ ∗
ϕ∗ α
ψ(U) U U U F(U)

• Let S = F(U) be an oriented subset of M. Then ψ(U) = ϕ(F(U)) = ϕ(S ) is

an oriented subset of W.

• The general pullback formula takes the form

Z Z
α= ϕ∗ α .
ϕ(S ) S

diffgeom.tex; April 12, 2006; 17:59; p. 94

94 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

• Remarks.

• We have only defined integrals of forms over subsets of M covered by a

single coordinate chart.

• We need to define the integrals over general submanifolds, covered by mul-

tiple charts.

diffgeom.tex; April 12, 2006; 17:59; p. 95

4.2. INTEGRATION OVER MANIFOLDS 95

4.2 Integration over Manifolds

4.2.1 Partition of Unity
• Let M be a manifold, p0 ∈ M be a point in M and (U, x) be a local coordinate
chart about p0 .
• Let xi = xi (p) be the local coordinates of the point p and x0i = xi (p0 ) be the
local coordinates of the point p0 .
• Let v
t n
X
||x − x0 || = (xi − x0i )2
i=1

• A neighborhood of p0 is a subset of M defined by

Bε (p0 ) = {p ∈ M | ||x − x0 || < ε} .

• Every neighborhood of a point in M is an open set in M.

• Let A ⊂ M be a subset of M. A point p ∈ M is called an accumulation
point (or a limit point) of A if every neighborhood of p contains at least
one point in A other than p.
• A subset of M is closed if and only if it contains all of its limit points.
• A closure of A, denoted by Ā, is a set obtained by adding to A all its
accumulation points.
• A closure of any set is a closed set.
• A function f : M → R is continuous if the inverse image of every open set
in R is open in M.
• Obviously, the set R \ {0} of non-zero real numbers is open.
• Thus, the subset of M where f is not equal to zero, that is, the set f −1 (R\{0}),
is open.
• The support of f is the closure of the set f −1 (R \ {0}),

supp f = f −1 (R \ {0}) .

diffgeom.tex; April 12, 2006; 17:59; p. 96

96 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

• Thus, f vanishes outside its support supp f .

• There could be points in supp f where f vanishes.

• A bump function is a smooth function ϕε : R → R such that

0 ≤ ϕε (t) ≤ 1 ∀t ∈ R

and
ε
 1, if |t| <


4

ϕε (t) = 

 0, if |t| >

 ε
2

• Similarly, we define support of any tensor field.

• For example, an n-form ω ∈ Λn on M defined by

ω = ϕε (||x − x0 ||) dx1 ∧ · · · ∧ dxn

has a support inside the neighborhood Bε (p0 ),

supp ω ⊂ Bε (p0 ) .

Such an n-form is called a bump form.

N
• Let {Uα }α=1 be a (finite) atlas for the manifold M.
N
• A partition of unity is a set {ϕα }α=1 of functions ϕα : M → R with the
properties

0 ≤ ϕα (p) ≤ 1 ∀α, p ∈ M
supp ϕα ⊂ Uα ∀α
XN
ϕα = 1
α=1

• Notice that for any α, supp ϕα is closed and ϕα vanishes outside Uα .

• A general theorem from analysis says that every manifold has a partition of
unity.

• Example.

diffgeom.tex; April 12, 2006; 17:59; p. 97

4.2. INTEGRATION OVER MANIFOLDS 97

4.2.2 Integration over Submanifolds

• A manifold M is compact if every open cover of M has a finite subcover.

• Thus, very compact manifold has a finite atlas.

• A subset of Rn is compact if it is closed and bounded.

• Let V be a p-dimensional compact oriented manifold.

N
• Let {Uα }α=1 be a finite atlas of V.

• Let each Uα be positively oriented.

• Let {ϕα } be a partition of unity on V.

• Let β be a p-form over V.

• The integral of β over V is defined by

Z N Z
X
β= ϕα β
V α=1 Uα

• This integral does not depend on the atlas and the partition of unity.

• Now, let V be a p-dimensional compact oriented submanifold of an n-

dimensional manifold M described by the inclusion map

i : V → M.

• Let β ∈ Λ p M be a p-form on M.

• The integral of a p-form β on M over V ⊂ M is defined by

Z Z
β= i∗ β .
V V

diffgeom.tex; April 12, 2006; 17:59; p. 98

98 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

4.2.3 Manifolds with boundary

• Recall that an open ball in Rn is the set

B (x0 ) = {x ∈ Rn | ||x − x0 || < ε}

• Let us consider also sets

H (x0 ) = {x ∈ Rn | ||x − x0 || < ε, xn − x0n ≥ 0} .

Such sets are called half-open.

• A n-dimensional manifold with boundary consists of the the interior M o

and the boundary ∂M.

• The interior M o is a genuine n-dimensional manifold such that all its points
have neighborhoods diffeomorphic to open balls in Rn .

• The boundary ∂M is a subset of M such that all its points have neighbor-
hoods diffeomorphic to half-open sets.

• Usually, the boundary ∂M is itself an (n − 1)-dimensional submanifold of

M (without boundary).

• Boundary may be disconnected. It can also be not smooth.

• Local coordinates (x1 , . . . , xn−1 , xn ) in M in the neighborhoods of points on

the boundary can always be chosen in such a way that (x1 , . . . , xn−1 ) are the
coordinates along the boundary and 0 ≤ xn < δ with some δ.

• A compact manifold is a manifold which is closed and bounded (say, as a

submanifold of some RN ).

• A closed manifold is a manifold which is compact and does not have a

boundary.

diffgeom.tex; April 12, 2006; 17:59; p. 99

4.3. STOKES’S THEOREM 99

4.3 Stokes’s Theorem

4.3.1 Orientation of the Boundary
• Let M be an n-dimensional orientable manifold with boundary ∂M, which
is an (n − 1)-dimensional manifold without boundary.
• Let M be oriented.
• Then an orientation on M naturally induces an orientation on ∂M.
• Let p ∈ ∂M and {e2 , . . . , en } be a basis in T p ∂M.
• Let N ∈ T p M be a tangent vector at p that is transverse to ∂M and points
out of M.
• Then {N, e2 , . . . , en } forms a basis in T p M.
• Then, by definition, the basis {e2 , . . . , en } has the same orientation as the
basis {N, e2 , . . . , en }. That is, {e2 , . . . , en } is positively oriented in ∂M if
{N, e2 , . . . , en } is positively oriented in M.

4.3.2 Stokes’ Theorem

Theorem 4.3.1 Let M be an n-dimensional manifold and V be a p-
dimensional compact oriented submanifold with boundary ∂V in M. Let
ω ∈ Λ p−1 M be a smooth (p − 1)-form in M. Then
•
Z Z
dω = ω.
V ∂V

Proof :
1. Let
i:V→M
be the inclusion map of the submanifold V.
2. Then Z Z Z
dω = i (dω) =
∗
d(i∗ ω)
V V
and Z Z
ω= i∗ ω
∂V ∂V

diffgeom.tex; April 12, 2006; 17:59; p. 100

100 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

3. Let
β = i∗ ω
4. Then we need to show
Z Z
dβ = β.
V ∂V

5. Let {Vα } be a finite cover of V and {ϕα } be the corresponding partition

of unity.
6. We have Z XZ
dβ = d(ϕα β)
V α Vα

and Z XZ
β= ϕα β .
∂V α ∂V

7. Thus, we need to show that

Z Z
d(ϕα β) = ϕα β .
Vα ∂V

8. The charts Vα can be of two different kinds: I) those that are fully in
the interior of V, that is, Vα ∩ ∂V = ∅, which are diffeomorphic to the
open balls in R p , and II) those which contain points of the boundary,
which are diffeomorphic to half-open balls in R p .
9. Case I. Let Vα be open sets in V fully in the interior of V.
10. Let Uα ⊂ R p be the open sets in R p such that fα : Uα → Vα be the
local coordinate diffeomorphisms.
11. Then Z Z Z
d(ϕα β) = fα∗ (d(ϕα β)) = d( fα∗ (ϕα β))
Vα Uα Uα

12. Let
γα = fα∗ (ϕα β) .
13. Then
p
X
γα = (−1)i−1 γα,i dx1 ∧ · · · ∧ dx
ci ∧ · · · ∧ dx p
i=1

diffgeom.tex; April 12, 2006; 17:59; p. 101

4.3. STOKES’S THEOREM 101

and p
X ∂γα,i
dγα = dx1 ∧ · · · ∧ dx p
i=1
∂xi
14. Therefore,
p Z p
∂γα,i 1 ∂γα,i 1
Z X X Z
dγα = dx ∧· · ·∧dx p
= dx ∧· · ·∧dx p = 0 .
Uα i=1 Uα ∂xi i=1 Rp ∂xi

15. Hence Z
d(ϕα β) = 0 .
Vα

16. Also, since Uα is disjoint from the boundary

Z
ϕα β = 0 .
∂V

17. Thus, for each chart disjoint from the boundary

Z Z
d(ϕα β) = ϕα β = 0 .
Vα ∂V

18. Case II. Now, let us consider the half-open charts Vα at the boundary.
19. Let Uα ⊂ R p be the half-open sets in R p such that fα : Uα → Vα be the
local coordinate diffeomorphisms.
20. Let
Wα = Vα ∩ ∂V
and
Yα = fα−1 (Wα ) .
21. Notice that for any point on Yα , x p = 0.
22. Then p Z
∂γα,i 1
Z X
d(ϕα β) = dx ∧ · · · ∧ dx p
p ∂x
i
Vα i=1 R

23. We have ∞
∂γα,i i
Z
dx = 0
−∞ ∂xi
for any i , p, and
∞
∂γα,p p
Z
dx = −γα,p .
0 ∂x p x p =0

diffgeom.tex; April 12, 2006; 17:59; p. 102

102 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

24. Therefore,
Z Z
d(ϕα β) = − γα,p (x1 , . . . , x p−1 , 0) dx1 · · · dx p
Vα R p−1

25. Further, Z Z Z Z
ϕα β = ϕα β = fα∗ (ϕα β) = γα
∂V Wα Yα Yα

26. Since on Yα , x p = 0, then dx p = 0. Therefore,

Z
γα = (−1) p−1 γα,p (x1 , . . . , x p−1 , 0)dx1 ∧ · · · ∧ dx p−1
Yα

27. We have that {∂1 , . . . , ∂ p } is positively oriented on V and −∂ p is the

outward normal to ∂V.
28. Thus, {∂1 , . . . , ∂ p−1 } has orientation (−1) p on ∂V.
29. Thus Z Z
ϕα β = − γα,p (x1 , . . . , x p−1 , 0)dx1 · · · dx p−1
∂V Yα

30. So, Z Z
d(ϕα β) = ϕα β
Vα ∂V
which establishes the theorem.

• Examples.
• Newton formula. Let C be an oriented curve, ∂C = Q − P be its boundary
and f be a smooth real-valued function. Then
Z Z
df = f = f (Q) − f (P)
C ∂C

• Green formula. Let S be an oriented surface in a 2-dimensional manifold

M with parameters (u1 , u2 ) and ∂S be its boundary with the induced ori-
entation. Let U be the preimage of S on the plane R2 with the boundary
∂U = [a, b]. Let A be a one-form in M. Then
∂A2 ∂A1 ∂(x1 , x2 ) 1
Z b
dx1 dx2
Z ! !
− 2 du ∧ du =
2
A1 + A2 dt
U ∂x ∂x ∂(u1 , u2 )
1 dt dt
a

diffgeom.tex; April 12, 2006; 17:59; p. 103

4.3. STOKES’S THEOREM 103

• Gauss formula. Let D be a region in a 3-dimensional manifold M and U

be the parameter preimage of D in R3 . Let ∂D be the boundary of D and ∂U
be the boundary of U. Let F be a 2-form in M. Then

∂F12 ∂F23 ∂F31 ∂(x1 , x2 , x3 ) 1

Z !
+ + du ∧ du2 ∧ du3
U ∂x 3 ∂x 1 ∂x 2 ∂(u 1 , u2 , u3 )

∂(x1 , x2 ) ∂(x2 , x3 ) ∂(x3 , x1 )

Z !
= F12 1 2 + F23 1 2 + F31 1 2 dz1 ∧ dz2
∂U ∂(z , z ) ∂(z , z ) ∂(z , z )

• Stokes’ formula. Let S be a surface in a 3-dimensional manifold M and

∂S be its boundary. Let U be the preimage of S in the parameter plane R2
and ∂U = [a, b] be its boundary. Let A be a 1-form in M. Then

∂A2 ∂A1 ∂(x1 , x2 ) ∂A3 ∂A1 ∂(x1 , x3 )

Z  ! !
− + −
U ∂x1 ∂x2 ∂(u1 , u2 ) ∂x1 ∂x3 ∂(u1 , u2 )
∂A3 ∂A2 ∂(x2 , x3 )
! )
+ − 3 du1 ∧ du2
∂x 2 ∂x ∂(u , u )
1 2
Z b
dx1 dx2 dx3
!
= A1 + A2 + A3 dt
a dt dt dt

diffgeom.tex; April 12, 2006; 17:59; p. 104

104 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

4.4 Poincaré Lemma

Definition 4.4.1 Let M be a manifold.
A p-form α on M is called closed if dα = 0.
•
A p-form α on M is called exact if there is a (p − 1)-form β such that
α = dβ. The form β is called a potential of α.

Theorem 4.4.1

1. Every exact form is closed (Poincaré Lemma).

That is, if α = dβ, then dα = 0.

2. The exterior product of closed forms is closed.

That is, if dα = 0 and dβ = 0, then d(α ∧ β) = 0.

3. The exterior product of a closed form and an exact form is exact.

That is, if dα = 0 and β = dγ, then there is σ such that α∧β = dσ.

• 4. Let M be an n-dimensional orientable compact manifold without

boundary and α be an exact n-form on M, that is, α = dβ for
some (n − 1)-form β. Then
Z
α = 0.
M

5. Let M be an n-dimensional oriented compact manifold with

boundary ∂M and α be a closed (n−1)-form on M, that is, dα = 0.
Then Z
α = 0.
∂M

Proof : Use Stokes theorem.



Definition 4.4.2 Let M be a manifold. Suppose that for every closed

oriented smooth curve C there is a smooth oriented 2-dimensional sur-
• face S and a map F : S → M such that ∂F(V) = C, that is, the curve C
is the boundary of the surface S . Then the manifold M is said to have
first Betti number equal to zero.

diffgeom.tex; April 12, 2006; 17:59; p. 105

4.4. POINCARÉ LEMMA 105

Theorem 4.4.2 Let M be a manifold with first Betti number equal to

• zero. Then every closed 1-form on M is exact. That is, if α is a 1-form
such that dα = 0, then there is a function f such that α = d f .
Proof :

1. Let α be a closed form on M.

2. Let x and y be two points in M and C xy be an oriented curve with the
initial point y and the final point x.
3. Let f be defined by Z
f (x) = α.
C xy

4. Then f is independent on the curve C xy and so is well defined.

5. Finally, we show that
df = α.

Theorem 4.4.3 Let α be a closed p-form in Rn . Then there is
(p − 1)-form β in Rn such that α = dβ.
•

That is, every closed form in Rn is exact.

Proof :

1. Let α be a closed p-form in Rn .

2. We define a (p − 1)-form β by
Z 1
βi1 ...i p−1 (x) = dτ τ p−1 x j α ji1 ...i p−1 (τx)
0

3. We can show that

dβ = α .

Corollary 4.4.1 Let M be a manifold and α be a closed p-form on
• M. Then for every point x in M there is a neighborhood U of x and a
(p − 1)-form β on M such that α = dβ in U.
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 106

106 CHAPTER 4. INTEGRATION OF DIFFERENTIAL FORMS

1. Use the fact that a sufficiently small neighborhood of a point in M is

diffeomorphic to an open ball in Rn .
2. Pullback the form α from M to Rn by the pullback F ∗ of the diffeo-
morphism F : V → U, where U ⊂ M and V ⊂ Rn .
3. Use the previous theorem.


diffgeom.tex; April 12, 2006; 17:59; p. 107

Chapter 5

Lie Derivative

5.1 Lie Derivative of a Vector Field

5.1.1 Lie Bracket
• Let M be a manifold.

• Let X be a vector field on M.

• Let ϕt : M → M be the flow generated by X.

• Let x ∈ M. Then ϕt (x) is the point on the integral curve of the vector field
X going through x and such that

ϕ0 (x) = x

and
dϕt (x)
= Xϕt (x) .
dt
• Let ϕt∗ : T x M → T ϕt (x) M be the differential of the diffeomorphism ϕt .

• Notice that
ϕ0 ∗ = Id
is the identity and
ϕ−t∗ = (ϕt∗ )−1
is the inverse transformation.

107
108 CHAPTER 5. LIE DERIVATIVE

• In local coordinates for small t we have

ϕit (x) = xi + tX i (x) + O(t2 ) ,

so that
∂ϕit (x) ∂X i (x)
(ϕt∗ ) j =
i
= δi
+ t + O(t2 )
∂x j j
∂x j

• Thus !i
d ∂X i (x)
ϕt∗ j = .
dt t=0
∂x j

• Let f be a smooth function on M.

• The flow ϕt naturally defines a new function

(ϕt∗ f )(x) = ( f ◦ ϕt )(x) = f (ϕt (x)) .

• Then for small t

f ◦ ϕ = f + tX( f ) + O(t2 ) .

• Thus
d
( f ◦ ϕt ) = X( f ) .
dt t=0

• Let Y be another vector field on M.

• Then
d
Y(( f ◦ ϕt )) = Y(X( f )) .
dt t=0

• Then at the point ϕt (x) we have two different well defined vectors, Yϕt (x) and
ϕt∗ Y x .

• Diagram.

Definition 5.1.1 A vector field Y is invariant under the flow ϕt gener-

ated by a vector field X if
•
Yϕt (x) = ϕt∗ Y x .

An invariant vector field Y is also called a Jacobi field.

diffgeom.tex; April 12, 2006; 17:59; p. 108

5.1. LIE DERIVATIVE OF A VECTOR FIELD 109

Definition 5.1.2 The Lie derivative of the vector field Y with respect
to the vector field X is the vector field LX Y defined at a point x by
•
1 
(LX Y) x = lim Yϕt (x) − ϕt∗ Y x
t→0 t

• Remark. Notice that this can also be written as

1 
(LX Y) x = lim ϕ−t∗ Yϕt (x) − Y x
t→0 t
or
d  
(LX Y) x = ϕ−t∗ Yϕt (x) .
dt t=0

Proposition 5.1.1
•
LX Y = [X, Y] .
Proof :

1. We compute in local coordinates

d  i
(LX Y)i = ϕ−t∗ Yϕt (x)
dt t=0
d h i
= (ϕ−t∗ ))i j Y j (ϕt (x))
dt t=0
d d
= (ϕ−t∗ ))i j Y j (x) + δij Y j (ϕt (x))
dt t=0
dt t=0
∂X i j ∂Y i j
= − Y (x) + X (x)
∂x j ∂x j
= [X, Y]i .

Definition 5.1.3 The Lie bracket of two vector fields X and Y is a

vector field [X, Y] such that for any smooth function f on M
•
[X, Y] = X(Y( f )) − Y(X( f )) .

• Notice that
[X, Y] = −[Y, X] .

diffgeom.tex; April 12, 2006; 17:59; p. 109

110 CHAPTER 5. LIE DERIVATIVE

• In particular,
LX X = 0 .

• In local coordinates the Lie bracket is given by

[X, Y]i = X j ∂ j Y i − Y j ∂ j X i .

• The linear ordinary differential equation

dY i (t)
[X, Y]i = − (∂ j X i )(ϕt (x))Y j (t) = 0
dt
for Y i (t) is called the Jacobi equation. For a given vector field X and given
initial conditions for Y i it defines a unique Jacobi field along the flow ϕt .

• In particular,
L∂i ∂ j = 0 .

5.1.2 Flow generated by the Lie Bracket

Theorem 5.1.1 Let M be a manifold. Let X and Y be vector fields on
M and ϕtX and ϕYt be the flows generated by X and Y respectively. Let
σt : M → M be a diffeomorphism defined by

σt = ϕY−t ◦ ϕ−t
X
◦ ϕYt ◦ ϕtX .

Let f be a smooth function on M. Then

• 1
[X, Y] x ( f ) = lim

2
f (σt (x)) − f (x)
t→0 t
d
= f (σ √t ) .
dt t=0

which means
d √
[X, Y] = σ .
dt t t=0
Proof :

1. Diagram.

diffgeom.tex; April 12, 2006; 17:59; p. 110

5.1. LIE DERIVATIVE OF A VECTOR FIELD 111

2. Use Taylor expansion in local coordinates.


Corollary 5.1.1 Let M be a manifold and W be a submanifold of M.
• Let X and Y be vector fields on M tangent to W. Then the Lie bracket
[X, Y] is also tangent to W.

diffgeom.tex; April 12, 2006; 17:59; p. 111

112 CHAPTER 5. LIE DERIVATIVE

5.2 Lie Derivative of Forms and Tensors

• Let X be a vector field on a manifold M.

• Let ϕt : M → M be the flow generated by X and ϕ∗t : T ϕt (x) M → T x M be

the corresponding pullback.

Definition 5.2.1 Let f be a function (0-form) on M. Then the Lie

derivative of f with respect to X is a function LX f defined by

• d ∗
(LX f ) x = (ϕ f ) x
dt t t=0
d
= f (ϕt (x)) .
dt t=0

Proposition 5.2.1 The Lie derivative of a function f with respect to a

vector field X is equal to
•
LX f = X( f ) .

• In local coordinates
LX f = X i ∂ i f .

Definition 5.2.2 Let α be a 1-form on M. The Lie derivative of α with

respect to X is a 1-form LX α defined by

• 1 ∗ 
(LX α) x = lim ϕt αϕt (x) − α x
t→0 t
d ∗

= ϕα
dt t x t=0

• We can immediately generalize this to p-forms.

Definition 5.2.3 Let α be a p-form on M. The Lie derivative of α with

respect to X is a p-form LX α defined by

• 1 ∗ 
(LX α) x = lim ϕt αϕt (x) − α x
t→0 t
d ∗

= ϕα
dt t x t=0

diffgeom.tex; April 12, 2006; 17:59; p. 112

5.2. LIE DERIVATIVE OF FORMS AND TENSORS 113

Proposition 5.2.2 The Lie derivative of a p-form α with respect to a

vector field X is given by
•
(LX α)i1 ...i p = X j ∂ j αi1 ...i p + α ji2 ...i p ∂i1 X j + · · · + αi1 ...i p−1 j ∂i p X j
Proof :
1. Use that
j
∂ϕtj1 (x) ∂ϕt p (x)
(ϕ∗t α)i1 ...i p (x) = ··· α j1 ... j p (ϕt (x))
∂xi1 ∂xi p

• In particular, for a 1-form α we have
(LX α)i = X j ∂ j αi + α j ∂i X j .

• More generally, since the flow ϕt : M → M is a diffeomorphism it naturally

acts on general tensor bundles of type (p, q), that is,
ϕ∗t : (T qp )ϕt (x) M → (T qp ) x M

For a general tensor field T of type (p, q) on M, ϕ∗t T is a tensor field of

type (p, q) defined by
j
k ...k ∂ϕtj1 (x) ∂ϕt q (x) ∂xk1 ∂xk p m ...m
(ϕ∗t T )i11...iqp (x) = · · · · · · T j11... jq p (ϕt (x))
∂x i 1 ∂x ∂ϕt (x)
i q m 1 mp
∂ϕt (x)

Definition 5.2.4 Let T be a tensor field of type (p, q) on M. The Lie

derivative of T with respect to X is a tensor field LX T of type (p, q)
defined by
• 1 ∗ 
(LX T ) x = lim ϕt T ϕt (x) − T x
t→0 t
d ∗

= ϕT
dt t x t=0

Proposition 5.2.3 The Lie derivative of a tensor field T of type (p, q)

with respect to a vector field X is given in local coordinates by
• k ...k k ...k k ...k k ...k
(LX T )i11...iqp = X j ∂ j T i11...iq p + T ji12 ...iqp ∂i1 X j + · · · + T i11...iq−1p j ∂iq X j
jk ...k k ...k j
−T i1 i22 ...iqp ∂ j X k1 − · · · − T i11...iq p−1 ∂ j X k p

diffgeom.tex; April 12, 2006; 17:59; p. 113

114 CHAPTER 5. LIE DERIVATIVE

Proof :

1. Use the definition of the action ϕ∗t .



• In particular, for a tensor gi j of type (0, 2) we obtain

(LX g)i j = X k ∂k gi j + gik ∂ j X k + gk j ∂i X k .

If gi j is a Riemannian metric, then this tensor is called the strain tensor.

5.2.1 Properties of Lie Derivative

Theorem 5.2.1 For any two tensors T and R and a vector field X the
Leibnitz rule holds
•
LX (T ⊗ R) = (LX T ) ⊗ R + T ⊗ (LX R)
Proof :

1. Use the Leibnitz rule for

d  ∗  d  ∗
ϕt (T ⊗ R) = (ϕt T ) ⊗ (ϕ∗t R)

dt dt

Theorem 5.2.2 Let α be a p-form, β be a q-form and X be a vector
field on M. Then the Leibnitz rule holds
•
LX (α ∧ β) = (LX α) ∧ β + α ∧ (LX β)
Proof :

1. Follows from the previous theorem and the definition of the wedge
product.

Theorem 5.2.3 For any 1-form ω and vector fields X and Y the Leib-
nitz rule holds
•
LX (ω(Y)) = (LX ω)(Y) + ω(LX Y)
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 114

5.2. LIE DERIVATIVE OF FORMS AND TENSORS 115

1. Computation in local coordinates.



• The Lie derivative of a 1-form ω with respect to a vector field X can be

defined in an intrinsic way. LX ω is a 1-form whose value on any vector field
Y is
(LX ω)(Y) = X(ω(Y)) − ω([X, Y]) .

• More generally, we have

Theorem 5.2.4 Let X and Y1 , . . . , Y p be vector fields on a manifold

M and α ∈ Λ p be a p-form. Then
 
LX α(Y1 , . . . , Y p ) = (LX α)(Y1 , . . . , Y p )
Xp
+ α(Y1 , . . . , LX Yi , . . . , Y p ) .
i=1

Proof : By induction or direct calculation.



• Remark. This can be used as an intrinsic definition of LX α.

Theorem 5.2.5 Let X and Y be any two vector fields and c ∈ R. Then

1. LX+Y = LX + LY ,

• 2. LcX = cLX ,

3. LX Y = −LY X,

4. [LX , LY ] = L[X,Y] .

Proof :

1.


diffgeom.tex; April 12, 2006; 17:59; p. 115

116 CHAPTER 5. LIE DERIVATIVE

Theorem 5.2.6 Let gi j be a Riemannian metric on an n-dimensional

manifold M and
vol = |g|dx1 ∧ · · · ∧ dxn
p

be the Riemannian volume form. Let X be a vector field on M. Then

• LX vol = ( div X)vol ,

where div X is a scalar function defined by

1 p 
div X = ∗LX vol = p ∂i |g|X i .
|g|
Proof :
1. Direct calculation. Use LX gi j .

• The scalar div X is called the divergence of the vector field X.
• Remark. Let Y1 , . . . , Yn be vector fields invariant under the vector field X.
Then
(LX vol )(Y1 , . . . , Yn )
div X =
vol (Y1 , . . . , Yn )
d
= log vol (Y1 , . . . , Yn ) .
dt t=0

• Thus, div X is the logarithmic rate of change of the volume along the flow.
• Remark. Let β ∈ Λn−1 be an (n − 1)-form defined by
β = iX vol .
In components
βi1 ...in−1 = X j |g|ε ji1 ...in−1
p

Then
dβ = ( div X)vol .
To prove this compute in local coordinates
 p 
(dβ)i1 ...in = n∂[i1 X j |g|ε ji2 ...in

diffgeom.tex; April 12, 2006; 17:59; p. 116

5.2. LIE DERIVATIVE OF FORMS AND TENSORS 117

• Recall that a linear map

A : Λ p → Λ p+r
is a derivation if r is even and for any α ∈ Λ p and β ∈ Λq
A(α ∧ β) = (Aα) ∧ β + α ∧ Aβ .
and an anti-derivation if r is odd and
A(α ∧ β) = (Aα) ∧ β + (−1) p α ∧ Aβ .

• Examples.
• The Lie derivative LX Λ p → Λ p is a derivation.
• The exterior derivative d : Λ p → Λ p+1 and the interior product iX : Λ p →
Λ p−1 are anti-derivations.
• Recall the following theorem about exterior derivative
Theorem 5.2.7 Let Y1 , . . . , Y p+1 be vector fields on a manifold M and
α ∈ Λ p be a p-form. Then
p+1
X  
(dα)(Y1 , . . . , Y p+1 ) = (−1)i+1 Yi α(Y1 , . . . , Ŷi , . . . , Y p+1 )
i=1
X
+ (−1) α([Yi , Y j ], . . . , Ŷi , . . . , Ŷ j , . . . , Y p+1 ) .
i+ j

1≤i< j≤p+1

• In particular, for p = 1 this takes the form

Theorem 5.2.8 Let X and Y be vector fields on a manifold M and
α ∈ Λ1 be a 1-form. Then

(dα)(X, Y) = X(α(Y)) − Y(α(X)) − α([X, Y]) .

Theorem 5.2.9 The Lie derivative commutes with the exterior deriva-
tive, that is,
LX d = dLX .
•
In other words, for any p-form α and a vector field X there holds

LX dα = dLX α
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 117

118 CHAPTER 5. LIE DERIVATIVE

1. Check for p = 0 and p = 1 explicitly.

2. Generalize for p > 1.

Theorem 5.2.10 (Cartan Formula) Let X be a vector field on a mani-
fold M and α ∈ Λ p be a p-form. Then

• LX α = iX dα + diX α .

That is,
LX = iX d + diX
Proof :

1. Verify it for p = 0 and p = 1.


Theorem 5.2.11 Let X and Y be vector fields on a manifold M and
α ∈ Λ p be a p-form. Then

• [LX , iY ]α = LX iY α − iY LX α = i[X,Y] α .

That is,
[LX , iY ] = i[X,Y] .
Proof : Exercise (Use Cartan formula).

Theorem 5.2.12 Let M be a n-dimensional manifold, X be a vector
field on M and ϕt : M → M be the corresponding flow of X. Let W be
a p-dimensional oriented compact submanifold of M and Wt = ϕt W be
• the image of W under the flow ϕt . Let α be a p-form on M. Then
Z Z
d
α= LX α
dt Wt Wt

diffgeom.tex; April 12, 2006; 17:59; p. 118

5.3. FROBENIUS THEOREM 119

5.3 Frobenius Theorem

5.3.1 Distributions
Definition 5.3.1 Let M be a n-dimensional manifold. A k-dimensional
distribution (or a tangent subbundle) ∆ : M → ∆ x ⊂ T x M is a smooth
assignment to each point x ∈ M a k dimensional subspace ∆ x of the
tangent space T x M.

An submanifold V of M that is everywhere tangent to the distribution is

called an integral manifold of the distribution.
•
A k-didmensional distribution ∆ is called integrable if at each point
x ∈ M there is a k-dimensional integral submanifold of ∆.

In other words, the distribution ∆ is integrable if everywhere in M there

exist local coordinates (x1 , . . . , xk , y1 , . . . , yn−k ) such that the coordinate
surfaces ya = ca , a = 1, . . . , n − k, ca being some constants, are integral
manifolds of the distribution ∆. Such a coordinate system is called a
Frobenius chart.

• Examples in R3 .

• A one-dimensional distribution in R3 is a family of lines at every point in

R3 (that is, a vector field). It is always integrable. Integral manifolds of the
distribution are families of integral curves of the vector field.

• A two-dimensional distribution is a smooth family of planes at every point

in R3 . It is integrable if there exist nonintersecting surfaces that are every-
where tangent to the planes (and fill up a region in R3 ). Not every two-
dimensional distribution is integrable!

• Let α be a 1-form in R3
α = αi (x)dxi ,
where i = 1, 2, 3. A two-dimensional distribution can be described by

α = 0.

• It is integrable if there exists a diffeomorphism xi = xi (t, u), u = (uµ ), µ =

1, 2, such that for a fixed t, xi = xi (t, u) describes a smooth one-parameter

diffgeom.tex; April 12, 2006; 17:59; p. 119

120 CHAPTER 5. LIE DERIVATIVE

family of surfaces S t and

∂xi
(F ∗ α)µ = αi (x(t, u)) =0
∂uµ
and for fixed u = (u1 , u2 ), xi = xi (t, u) describes a smooth two-parameter
family of curves Cu transversal to the surfaces S t .

• Let t = t(x) and uµ = uµ (x) be the inverse diffeomorphism. Then the level
surfaces t(x) = C are the integral surfaces S t of the distribution, so that the
coordinate system (t, u1 , u2 ) is the Frobenius chart.

• Therefore, we have

∂xi µ ∂xi
α = αi du + α i dt
∂uµ ∂t
∂xi
= αi dt
∂t
= f dt ,

where
∂xi
f = αi .
∂t
• Therefore,
dα = d f ∧ dt .
and
α ∧ dα = 0 .

• Thus the distribution is integrable if

α ∧ dα = 0 .

This is called the Euler’s integrability condition.

• In local coordinates
α[i ∂ j αk] = 0 .
In R3 this means
α · curl α = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 120

5.3. FROBENIUS THEOREM 121

Definition 5.3.2 Let ∆ be a distribution on M. It is said to be in

involution if it is closed under Lie brackets, that is, for any two vector
• field X and Y in ∆ the Lie bracket [X, Y] is also in the distribution, or

[∆, ∆] ⊂ ∆ .

• Let ∆ be an integrable distribution. Let X and Y be two vector fields in ∆.

Then X and Y are tangent to the integral manifold of ∆. Therefore, the Lie
bracket [X, Y] is also tangent to the integral manifold and is in ∆.

• Proposition 5.3.1 Every integral distribution is in involution.

Definition 5.3.3 Let α be a 1-form on M. Let x ∈ M be a point such

that α x , 0. The null space of the form α at x is the (n − 1)-dimensional
• subspace of T x M spanned by the vectors X ∈ T x M such that

α(X) = 0 .

• Remark. A 1-form is also called a Pfaffian.

• Let α1 , . . . , αn−k be (n − k) linearly independent 1-forms such that

α1 ∧ · · · ∧ αn−k , 0

• Let N1 , . . . Nn−k be their null spaces. Then the intersection of the null spaces
forms a k-dimensional distribution ∆
n−k
\
∆= Nµ .
µ=1

• Locally this distribution is described by (n − k) Pfaffian equations

α1 = · · · = αn−k = 0 .

• Remarks.

• Let ∆ be in involution. Then dαµ |∆ = 0, µ = 1, . . . , (n − k), that is, for any

X, Y ∈ ∆
(dαµ )(X, Y) = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 121

122 CHAPTER 5. LIE DERIVATIVE

• Now, suppose that dαµ |∆ = 0, µ = 1, . . . , (n − k). Then the distribution ∆ is

in involution.
Theorem 5.3.1 Let M be a n-dimensional manifold and α1 , . . . , αn−k
be (n − k) one-forms and ω be a (n − k)-form defined by

ω = α1 ∧ · · · ∧ αn−k .

Suppose that the forms α1 , . . . , αn−k are linearly independent and ω , 0.

Let ∆ be the k-dimensional distribution defined by the intersection of the
null spaces of the 1-forms α1 , . . . , αn−k . Then the following conditions
are locally equivalent

1. the distribution ∆ is in involution,

•
2. dαµ |∆ = 0, that is, for any X, Y ∈ ∆

(dαµ )(X, Y) = 0 .

3. dαµ ∧ ω = 0,

4. there exist 1-forms γµν such that

n−k
X
dαµ = γµν ∧ αν .
ν=1

Proof :
1. (1) ⇔ (2). Use the formula
dα(X, Y) = X(α(Y)) − Y(α(X)) − α([X, Y]) .

2. (4) =⇒ (2) and (4) =⇒ (3). Trivial.

3. (2) =⇒ (4). Suppose dαµ |∆ = 0.
4. Let β1 , . . . , βk be 1-forms such that α1 , . . . , αn−k , β1 , . . . , βk is a basis in
T x∗ M.
5. Let e1 , . . . en−k , v1 , . . . , vk be the dual basis in T x M.
6. Then for µ = 1, . . . , (n − k), i = 1, . . . , k
αµ (vi ) = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 122

5.3. FROBENIUS THEOREM 123

Thus, span {vi } = ∆.

7. We have
X X X
dαµ = Cµλν αλ ∧ αν + Aµλi βi ∧ αν + Bµi j βi ∧ β j
µ<ν i,ν i< j
X X
= γµν ∧ αν + Bµi j βi ∧ β j .
ν i< j

8. Thus,
Bµi j = (dαµ )(vi , v j ) = 0 .
9. (3) =⇒ (4). Suppose that dαµ ∧ ω = 0.
10. Then we have
X X
0 = dαµ ∧ ω = γµν ∧ αν ∧ ω + Bµi j βi ∧ β j ∧ ω .
ν i< j
X
= Bµi j βi ∧ β j ∧ α1 ∧ · · · ∧ αn−k .
i< j

11. Thus, Bµi j = 0.


• Remarks.
• A k-dimensional distribution ∆ can be described locally by either k linearly
independent vector fields that span ∆ or by (n − k) linearly independent
1-forms whose common null space is ∆.
• If dαµ = ν=1 γµν ∧ αν for some 1-forms γµν , then we write
Pn−k

dαµ = 0 (mod α) .

5.3.2 Frobenius Theorem

Definition 5.3.4 Let M be an n-dimensional manifold, W be a k-
dimensional manifold and F : W → M be a smooth map.

Then F is an immersion if for each x ∈ W the differential F∗ : T x W →

•
T F(x) M is injective, that is, Ker F∗ = 0.

The image F(W) of the manifold W is called an immersed submani-

fold.

diffgeom.tex; April 12, 2006; 17:59; p. 123

124 CHAPTER 5. LIE DERIVATIVE

• Let M be an n-dimensional manifold.

• Let ∆ be a k-dimensional distribution on M.

• Let X1 , . . . , Xk be vector fields that span ∆ and ϕ1t , . . . , ϕkt be the correspond-
ing flows.

• Let x ∈ M be a point in M.

• Let B ∈ Rk be a sufficiently small open ball in Rk around the origin.

• We define
F:B→M
for any t = (t1 , . . . , tk ) ∈ B by
 
F(t) = ϕktk ◦ · · · ◦ ϕ1t1 (x) .

• Note that F(0) = x.

• Then for the differential at the origin

F ∗ T 0 B = Rk → T x M

we have
∂
F∗ = Xµ x ,
∂tµ t=0

where µ = 1, . . . , k.

• Thus,
F∗ T 0 B = ∆ x .

• So, F∗ is injective at the origin and, hence, in some neighborhood of the

origin.

• Therefore, F(B) is tangent to ∆ at the single point x.

• Thus, F(B) is an immersed submanifold of M.

diffgeom.tex; April 12, 2006; 17:59; p. 124

5.3. FROBENIUS THEOREM 125

Theorem 5.3.2 Frobenius Theorem. Let M be an n-dimensional man-

ifold, ∆ be a k-dimensional distribution in involution on M, X1 , . . . , Xk
be vector fields that span ∆ and ϕ1t , . . . , ϕkt be the corresponding flows.
Let x ∈ M, B be a sufficiently small ball around the origin in Rk and
F : B → M be defined by
 
F(t) = ϕktk ◦ · · · ◦ ϕ1t1 (x) .
•
Then:

1. F(B) is an immersed submanifold of M,

2. F(B) is an integral manifold of ∆,

3. the distribution ∆ is integrable.

Proof :

1. (I) done earlier.

2. (II). We need to show that ∆ is tangent to F(B) at each point of F(B).
3. We have for µ = 1, . . . , k
∂  µ
 h  i
F∗ = ϕ k
◦ · · · ◦ ϕ Xµ ϕk−1
◦ · · · ◦ ϕt1 (x) ,
1
∂tµ tk ∗ tµ ∗ tk−1

4. Thus, the tangent space T F(t) F(B) has a basis

   
ϕktk ∗ ◦ · · · ◦ ϕ2t2 ∗ X1 ϕ1t1 (x)
  h  i
ϕktk ∗ ◦ · · · ◦ ϕ3t3 ∗ X2 ϕ2t2 ◦ ϕ1t1 (x)
...
h  i
ϕktk ∗ Xk−1 ϕk−1 tk−1 ◦ · · · ◦ ϕ1
t1 (x)
h  i
Xk ϕktk ◦ · · · ◦ ϕ1t1 (x) .

5. Therefore, we need to show that for each µ = 1, . . . , k, the differentials

ϕµt∗ map the distribution ∆ into itself, that is,

ϕµt∗ (∆) ⊂ ∆ .

6. Claim: This follows from the fact that ∆ is in involution.

diffgeom.tex; April 12, 2006; 17:59; p. 125

126 CHAPTER 5. LIE DERIVATIVE

7. Let y ∈ F(B) and Y ∈ ∆y .

8. Let Yµt = ϕµt∗ Y, µ = 1, . . . , k.
9. We will show that Yµt ∈ ∆ϕµt (y) for µ = 1, . . . , k.
10. Let ∆ be defined by the 1-forms
α1 = · · · = αn−k = 0 .

11. The vector field Yµt is invariant under the flow of Xµ , so along the
orbits ϕµt (y) we have
LXµ Yµt = 0 .
µ
12. Let f1µ , . . . , fn−k be real valued functions defined by
fiµ (t) = αi (Yµt ) , (i = 1, . . . , n − k) .

13. Then at t = 0 we have the initial conditions

fi (0) = αi (Y) = 0 .

14. Further,
d µ
f (t) = iYµt iXµ dαi .
dt i
15. Since ∆ is in involution, by using dαi = k βik ∧ αk , we obtain
P

n−k
d µ
γi j (Xµ ) f jµ (t) .
X
fi (t) =
dt j=1

16. Thus, the above system of the differential equations has the unique
solution
fiµ (t) = 0 .
17. Thus, Yµt is in ∆ for all t and ∆ is tangent to the immersed ball F(B)
at each point of F(B).
18. (III). By constructing Frobenius chart.
19. That is, we construct coordinates x1 , . . . , xk , y1 , . . . , yn−k so that the im-
mersed balls F(B) are described locally by
y1 = c1 , . . . , yn−k = cn−k ,
where ci are constants.

diffgeom.tex; April 12, 2006; 17:59; p. 126

5.3. FROBENIUS THEOREM 127

20. We define a transversal to F(B) (n − k)-dimensional submanifold W

with local coordinates y1 , . . . , yn−k .
21. If the integral balls are sufficiently small, then for different points of
W the integral balls at those points are disjoint.


5.3.3 Foliations
• Let M be an n-dimensional manifold and ∆ be a k-dimensional distribution
on M.

• Let ∆ be in involution, and, therefore, integrable.

• Then, at each point x of M there exists an integral manifold of ∆.

• The integral manifold may return to the Frobenius coordinate patch around
the point x infinitely many times.

Definition 5.3.5 The integral manifolds of an integrable distribution

define a foliation of M.

• Each connected integral manifold is called a leaf of the foliation.

A leaf that is not properly contained in another leaf is called a maximal

leaf.
• A maximal leaf is not necessarily an embedded submanifold.

• An immersed submanifold does not have to be an embedded submanifold.

Theorem 5.3.3 A maximal leaf of a foliated manifold is an immersed
submanifold.
•
More precisely, for each maximal leaf V there is an injective immersion
F : V → M that realizes V globally.

• Examples.

diffgeom.tex; April 12, 2006; 17:59; p. 127

128 CHAPTER 5. LIE DERIVATIVE

5.4 Degree of a Map

5.4.1 Gauss-Bonnet Theorem
• Let us consider a compact oriented two-dimensional manifold M (a surface
embedded in R3 ).

• Let xi , i = 1, 2, 3, be the Cartesian coordinates in R3 and uµ , µ = 1, 2, be the

local coordinates on M.

• Let F : M → R3 be the embedding map defined locally by

xi = F i (u) .

• The differential of the map F is given by the matrix

∂xi
(F∗ )i µ = .
∂uµ

• We assume that F∗ is onto, that is, rank F∗ = 2.

• The tangent space T x M is spanned by the vectors eµ , µ = 1, 2, with compo-

nents
∂xi
eiµ = µ .
∂u
• Let δi j be the Euclidean metric in R3 .

• Then the induced metric on M is defined by

gµν = (eµ , eν ) = δi j eiµ eνj .

This matrix is also called the first fundamental form. It describes the
intrinsic geometry of the surface.

• Let N be the vector in R3

N = e1 × e2
with components
∂x j ∂xk
Ni = εi jk e1j ek2 = εi jk .
∂u1 ∂u2

diffgeom.tex; April 12, 2006; 17:59; p. 128

5.4. DEGREE OF A MAP 129

• Then N is a vector field that is everywhere normal to M.

• Notice that the norm of the normal vector is

||N||2 = ||e1 ||2 ||e2 ||2 − (e1 , e2 )2 .

• The second fundamental form, or the extrinsic curvature is defined by

the matrix
1 ∂eµ 1 ∂2 x i
!
bµν = , N = Ni .
||N|| ∂uν ||N|| ∂uµ ∂uν

• The second fundamental form describes the extrinsic geometry of the sur-
face M.

• The mean curvature of M is defined by

H = gµν bµν .

• The Gauss curvature of M is defined by

det bµν
K= .
det gαβ

• Gauss has shown that the K is an intrinsic invariant. In fact,

1
K = R12 12 = R ,
2

where R12 12 is the only non-vanishing components of the Riemann curvature

of the metric g and R is the scalar curvature. This will be discussed later.

• The Gauss map is the map ϕ : M → S 2 from M to S 2 defined by

N(x)
ϕ(x) = ,
||N(x)||

that is, it associates to every point x in M the unit normal vector at that
point.

diffgeom.tex; April 12, 2006; 17:59; p. 129

130 CHAPTER 5. LIE DERIVATIVE

Theorem 5.4.1 Let M be a closed (compact without boundary) ori-

ented 2-dimensional surfacepembedded in R3 . Let gµν be the induced
Riemannian metric, dvol = |g|dx be the Riemannian volume element
on M, K be the Gaussian curvature of M and ϕ : M → S 2 be the Gauss
normal map. Then
• Z
1
dvol K = deg(ϕ)
4π M
is an integer that does not depend on the metric and does not change
under smooth deformations of the surface.
Proof : Later.

• Remark. For a 2-surface of genus g (with g holes)

deg(ϕ) = 1 − g .

• The Euler characteristic of the surface M is a topological invariant equal

to
χ = 1 − g.

5.4.2 Laplacian
• Let gi j be a Riemannian metric on a manifold V. Let h be a function and X
be its gradient vector field defined by

X i = gi j ∂ j h .

• The Laplacian operator ∆ on the scalar function h is defined by

∆h = div X .

• In local coordinates
∆h = g−1/2 ∂i (g1/2 gi j ∂ j h) ,
where g = det gi j .

• The operator (−∆) : C ∞ (V) → C ∞ (V) acting on smooth functions on a

compact manifold V without boundary can be extended to a self-adjoint
operator on the Hilbert space L2 (V).

diffgeom.tex; April 12, 2006; 17:59; p. 130

5.4. DEGREE OF A MAP 131

• It has a discrete non-negative real spectrum {λ}∞ k=0 bounded from below by
zero
0 = λ0 ≤ λ1 ≤ λ2 · · ·
with finite multiplicities.

• The eigenfunctions {hk }∞ 2

k=0 form an orthonormal basis in L (V), that is,
Z
(hk , hl ) = dvol hk hl = δkl .
V

• For each f ∈ L2 (V) there is a Fourier series

∞
X
f = ak hk ,
k=0

where Z
ak = (hk , f ) = f hk .
V

• The lowest eigenvalue is 0. It is simple (has multiplicity 1). The corre-

sponding eigenfunction is the constant h0 = [vol (V)]−1/2 .

Lemma 5.4.1 Let f be a function on a closed manifold M such that

= Let {λk , hk }∞
R
M
f 0. k=1 be the spectral resolution of the operator (−∆).
Then the equation
∆h = f
has a unique solution given by the Fourier series
• ∞
X 1
h=− ak hk ,
k=1
λ k

where Z
ak = (hk , f ) = f hk .
M

Proof :

1.


diffgeom.tex; April 12, 2006; 17:59; p. 131

132 CHAPTER 5. LIE DERIVATIVE

Lemma 5.4.2 Let V be a closed (compact without boundary) oriented

n-dimensional manifold and ω be an n-form on V. Then ω is exact if
• and only if Z
ω = 0.
V

Proof :

1. (I). If ω is exact, then ω = dσ for some (n − 1)-form σ.

2. Then by Stokes theorem Z
ω = 0.
V

3. (II). Suppose that Z

ω = 0.
V

4. Let us fix a Riemannian metric on V and let vol be a Riemannian

volume element.
5. Then
ω = f vol
for some function f which satisfies
Z
f = 0.
V

6. Let h be the solution of the equation

∆h = f .

7. Let X be the gradient of the function h.

8. We define an (n − 1)-form σ by

σ = iX vol .

9. Then
dσ = ω .


diffgeom.tex; April 12, 2006; 17:59; p. 132

5.4. DEGREE OF A MAP 133

5.4.3 Brouwer Degree

• Let M and V be two closed (compact without boundary) oriented n-dimensional
manifolds.

• Let ϕ : M → V be a smooth map.

• Let ω ∈ Λn V be an n-form on V.

• Suppose that Z
ω , 0.
V
Then we can normalize it so that
Z
ω = 1.
V

• Then we can consider the quantity

ϕ∗ ω
R
M
,
ω
R
V

which can also be written as

ω
R
ϕ(M)
.
ω
R
V

• This quantity counts how many times the image of M wraps around V.

Corollary 5.4.1 Let M and V be n-dimensional manifolds and ϕ :

RM → V be aR smooth map. Let α and β be n-forms on V such that
V
α , 0 and V β , 0. Then
•
ϕ α ϕ∗ β
R R
∗
M
= RM
.
α β
R
V V

Proof :

1. Let  
 α β
ω =  R − R  .


V
α V
β

diffgeom.tex; April 12, 2006; 17:59; p. 133

134 CHAPTER 5. LIE DERIVATIVE

2. We have Z
ω = 0.
V

3. Therefore, the form ω is exact.

4. Hence, the form ϕ∗ ω is exact.
5. Thus, Z
ϕ∗ ω = 0 .
M


• The quantity
ϕ∗ ω
R
deg(ϕ) = R M

V
ϕ
does not depend on the choice of the form ω but only on the map ϕ. It is
called the Brouwer degree of the map ϕ.

• Example.

• In the case of one-dimensional manifolds the degree of the map ϕ : M → S 1

is called the winding number.

• Picture.
Theorem 5.4.2 Let V and M be n-dimensional compact oriented man-
ifolds without boundary. Let ϕ : M → V be a smooth map. Let y ∈ V
be a regular value of ϕ so that the differential ϕ∗ : T x M → T y V at any
point x ∈ ϕ−1 (y) is bijective (isomorphism). Then
• X
deg(ϕ) = sign (ϕ(x)) ,
x∈ϕ−1 (y)

where
sign (ϕ(x)) = sign (det(ϕ∗ )) .
Proof :

1. (I). Claim: the preimage ϕ−1 (y) of a regular value is a finite set, that is,

ϕ−1 (y) = {xi ∈ M | ϕ(xi ) = y, i = 1, 2, . . . , N} .

diffgeom.tex; April 12, 2006; 17:59; p. 134

5.4. DEGREE OF A MAP 135

2. Suppose ϕ−1 (y) is infinite.

3. Then by compactness argument ϕ−1 (y) has a limit point x0 ∈ M, which
is a regular point.
Indeed, every sequence has a convergent subsequence. Thus, there is a
sequence (xk ), such that xk ∈ ϕ−1 (y), converging to some x0 , xk → x0 ,
so that ϕ(x0 ) = y. Thus x0 is a regular point of M.
4. Since ϕ∗ : T x0 M → T y V is bijective at x0 , the map ϕ : M → V is a
diffeomorphism in a neighborhood of x0 .
That is, det ϕ∗ | x0 , 0.
5. This contradicts the fact that there is sequence of points xk → x0 such
that ϕ(xk ) = y.
Since, otherwise there exist infinitely many points xk ∈ M such that
ϕ(xk ) = ϕ(x0 ) = y contradicting the fact that ϕ is bijective.
6. (II). Claim: The point y ∈ V has a neighborhood whose inverse image
is a disjoint union of neighborhoods of the preimages of y, each of
which is diffeomorphic to Vy .
7. Let Wi be disjoint neighborhoods of xi ∈ M such that ϕ : Wi → Vi =
ϕ(Wi ) are diffeomorphisms.
8. Let S = ϕ(M − ∪i=1
N
Wi ).
N
9. Since M − ∪i=1 Wi is compact, then S is compact (and closed in V).
10. Let O = V − S .
11. Then O is open and is a neighborhood of y.
12. Now, we define
Vy = O ∩ ∩i=1
N
Vi .

13. Then
Vy ⊂ O, Vy ⊂ Vi .

14. Let
Ui = ϕ−1 (O) ∩ Wi .

15. Then
U i ⊂ Wi ⊂ M .

diffgeom.tex; April 12, 2006; 17:59; p. 135

136 CHAPTER 5. LIE DERIVATIVE

16. Then
ϕ−1 (Vy ) = ∪i=1
N
Ui ,
and ϕ : Ui → Vy are diffeomorphisms.
17. (III). Let ω be an n-form on V with support in Vy such that ω = 1.
R
V
18. Let yi , i = 1, . . . , n, be local coordinates in Vy .
19. Since each ϕ : Ui → Vy are diffeomorphisms, one can use yi as local
coordinates on Ui .
20. In such coordinates the diffeomorphism ϕ : Ui → Vy is the identity
map.
21. We notice that the orientation of Ui is described by sign ϕ(xi ).
22. Thus
Z N Z
X N Z
X
deg(ϕ) = ϕω=
∗
ϕω=
∗
ω
M i=1 Ui i=1 ϕ(Ui )

N
X Z N
X
= sign ϕ(xi ) ω= sign (ϕ(xi )) .
i=1 Vy i=1

Corollary 5.4.2

1. The Brouwer degree deg(ϕ) of any map ϕ : M → V is an integer.

2. The sum X
• sign (ϕ(x))
x∈ϕ−1 (y)

is independent on the regular value y ∈ V.

3. The Brouwer degree deg(ϕ) remains constant under continuous

deformations of the map ϕ.

• Problem. Let ω be the volume form on S n in Rn+1 given by

n+1
X
ω= ck ∧ · · · ∧ dxn+1 .
(−1)k dx1 ∧ · · · ∧ dx
k=1

Show that the antipodal map ϕ : S n → S n has degree deg(ϕ) = (−1)n+1 .

diffgeom.tex; April 12, 2006; 17:59; p. 136

5.4. DEGREE OF A MAP 137

• Problem. Let M be a closed oriented n-dimensional manifold and ϕ : M →

S n be a smooth map. We identify ϕ(x) with a unit vector field v on S n in
Rn+1 . Let vol be the volume (n+1)-form in Rn+1 . Let vol (S n ) be the volume
of the unit sphere S n . Let uµ , µ = 1, . . . , n, be local coordinates on M. Show
that

∂v ∂v
Z !
1
deg(ϕ) = vol v, 1 , · · · , n du1 ∧ · · · ∧ dun
vol (S n ) M ∂u ∂u

5.4.4 Index of a Vector Field

• Let M be a closed (compact without boundary) n-dimensional submanifold
of Rn+1 that is a boundary of a compact region U ⊂ Rn+1 , that is,

M = ∂U .

• Let N be an outward-pointing normal to M.

• Then N induces an orientation on M.

• Let v be a unit vector field on M.

• Let S n be the unit n-sphere embedded in Rn centered at the origin.

• We identify the unit vectors in Rn+1 with points in S n .

• Let ϕ : M → S n be a map defined for every x ∈ M by

ϕ(x) = v(x) .

• The Kronecker index of the vector field v on M is defined as the degree of

the map ϕ:
Index(v) = deg(ϕ) .

• In general, if v is a nonvanishing vector field on M, then its index is defined

as the Kronecker index of the unit vector field v/||v||.

• Example. M = S 1 in R2 .

diffgeom.tex; April 12, 2006; 17:59; p. 137

138 CHAPTER 5. LIE DERIVATIVE

• Problem. Let M = ∂U, where U ⊂ Rn+1 is a compact region as above.

Show that if a vector field v on M can be extended to a nonvanishing vector
field on all of the interior region U, then Index(v) = 0.
In other words, a vector field on M with a non-zero index has a singularity
(that is, it vanishes) at least at one point inside M.
• Problem. Suppose that a unit vector field v in Rn+1 is such that it is smooth
in U except for a finite number of points xa , a = 1, . . . , N. Let Ba be suf-
ficiently small balls around the points xa so that they are in U. Then v is
non-vanishing in U \ ∪a=1N
Ba . Then ∂Ba are small spheres with outward-
poiting normals. Let Index(v|∂Ba ) be the indices of v on ∂Ba . Show that
N
X
Index(v) = Index(v|∂Ba ) .
a=1

That is, the index of a vector field on a closed manifold M is equal to the
sum of indices inside M.
Theorem 5.4.3 Let vt , t ∈ [0, 1], be a smooth family of non-vanishing
vector field on a closed manifold M. Then
•
Index(v0 ) = Index(v1 ) .
Proof :
1. Follows from the fact that the index is an integer.

• Problem. Show that if a vector field v on a sphere S n in Rn+1 never points
to the origin, then
Index(v) = 1 .

• Corollary. The index of every non-vanishing vector field tangent to S n is

equal to
Index(v) = 1 .

Theorem 5.4.4 Brouwer Fixed Point Theorem. Let B be a closed

• unit ball in Rn+1 centered at the origin. Let ϕ : B → B be a smooth
map. Then ϕ has a fixed point.
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 138

5.4. DEGREE OF A MAP 139

1. Let v be a vector field in Rn+1 defined by

v(x) = ϕ(x) − x .

2. Then v never points to the origin. So, Index(v) = 1, and, therefore, ϕ

has a fixed point in B.
3. Alternatively, suppose that ϕ does not have a fixed point.
4. Then v is non-vanishing.
5. Let ψ : B → ∂B = S n be a map defined as follows. The point ψ(x) is
the point on the sphere S n where the line from the ϕ(x) to the point x
intersects S n .
6. Then ψ(x) = x for any x ∈ S n .
7. Let α be an n-form on S n normalized by α = 1.
R
Sn
8. Since ψ is an identity map on S n , the form ψ∗ α is an n-form on B
whose restriction to S n is equal to α.
9. Since also S n = ∂B and dα = 0, we have
Z Z Z
1= α= ψα=
∗
d(ψ∗ α) = 0 ,
Sn ∂B B

which is a contradiction.


• Problem. Let M be a closed n-dimensional submanifold of Rn+1 Let v be

a unit vector field on M. Let vol be the volume (n + 1)-form in Rn+1 . Let
vol (S n ) be the volume of the unit sphere S n . Let uµ , µ = 1, . . . , n, be local
coordinates on M. Show that

∂v ∂v
Z !
1
Index(v) = vol v, 1 , · · · , n du1 ∧ · · · ∧ dun
n
vol (S ) M ∂u ∂u

• Problem. If v is a non-vanishing vector field, not necessarily unit, then

∂v ∂v
Z !
1 1
Index(v) = vol v, 1 , · · · , n du1 ∧ · · · ∧ dun
vol (S n ) M ||v||n+1 ∂u ∂u

diffgeom.tex; April 12, 2006; 17:59; p. 139

140 CHAPTER 5. LIE DERIVATIVE

Corollary 5.4.3 Let M be a closed n-dimensional submanifold of Rn+1

such that M is the boundary of a compact region U ⊂ Rn+1 . Let fi ,
i = 1, . . . , (n + 1), be smooth functions on U. Let
n+1
X
|| f || =
2
| fi |2 .
i=1

Suppose the functions fi do not have common zeros on M, that is, || f || ,

0 on M. Let uµ , µ = 1, . . . , n, be local coordinates on M. Let
∂ fi ∂ fi
det( f, d f ) = εi1 i2 ...in in+1 fi1 12 · · · n+1
∂u ∂un
 ∂ f1 ∂ f1 
•  f1 ∂u1 · · · ∂u n 
 .. . . .

= det  . .
. . . .
.


 ∂ fn ∂ fn 
fn ∂u1 · · · ∂un

Suppose that
Z
1
n+1
det( f, d f )du1 ∧ · · · ∧ dun , 0 .
M || f ||
Then the functions fi have a common zero in U, that is, the system of
(n + 1) equations
f1 = · · · = fn+1 = 0
has a solution in U.
Proof : Follows from above.


5.4.5 Linking Number

• Let Cµ : S 1 → R3 , µ = 1, 2, be two nonintersecting smooth closed curves
(loops) in R3 described by

x = x1 (θ1 ) , x = x2 (θ2 ) .

• Let T 2 = S 1 × S 1 be the torus with the local coordinates θ1 , θ2 .

diffgeom.tex; April 12, 2006; 17:59; p. 140

5.4. DEGREE OF A MAP 141

• Let ϕ : T → S 2 be a smooth map defined by

x1 (θ1 ) − x2 (θ2 )
ϕ(θ) = .
||x1 (θ1 ) − x2 (θ2 )||

• The Gauss linking number of the loops C1 and C2 is defined by

Link(C1 , C2 ) = deg(ϕ) .

• Problem. Let
x12 = x2 − x1 .
Show that
Z Z !
1 1
Link(C1 , C2 ) = x12 × dx12 · dx1
4π C1 C2 ||x12 ||3
Z 2π Z 2π
1 {[x2 (θ2 ) − x1 (θ1 )] × ẋ2 (θ2 )} · ẋ1 (θ1 )
= dθ1 dθ2
4π 0 0 ||x2 (θ2 ) − x1 (θ1 )||3

• Let V be an orientable surface in R3 such that ∂V = C1 .

• Let N be the normal vector to V consistent with the orientation of C1 .

• Let the curve C2 intersect V transversally.

• The intersection number V ◦ C2 of the curve C2 and the surface V is the

signed number of intersections of C2 and V, with an intersection being pos-
itive if the tangent vector to C2 at the point of the intersection has the same
direction as N, that is,
X
V ◦ C2 = sign (N, ẋ) x ,
i
xi ∈V

where the sum goes over all intersection points xi .

• Problem. Show that

Link(C1 , C2 ) = V ◦ C2 .

diffgeom.tex; April 12, 2006; 17:59; p. 141

142 CHAPTER 5. LIE DERIVATIVE

diffgeom.tex; April 12, 2006; 17:59; p. 142

Chapter 6

Connection and Curvature

6.1 Affine Connection

6.1.1 Covariant Derivative
Definition 6.1.1 Let M be an n-dimensional manifold. An affine con-
nection is an operator

∇ : C ∞ (T M) × C ∞ (T M) → C ∞ (T M)

that assigns to two vector fields X and Y a new vector field ∇X Y, that
is linear in both variables, that is, for any a, b ∈ R and any vector fields
X, Y and Z,
•
∇X (aY + bZ) = a∇X Y + b∇X Z
∇aX+bY Z = a∇X Z + b∇Y Z

and satisfies the Leibnitz rule, that is, for any smooth function f ∈
C ∞ (M) and any two vector fields X and Y,

∇X ( f Y) = X( f )Y + f ∇X Y
= (d f )(X)Y + f ∇X Y

• Let xµ , µ = 1, . . . , n, be local coordinates and ∂µ be the basis of vector fields.

It will be called a coordinate frame for the tangent bundle.

• A basis of vector fields ei = eµi ∂µ , i = 1, . . . , n, for the tangent bundle T M is

143
144 CHAPTER 6. CONNECTION AND CURVATURE

called a frame.

• We will denote partial derivatives by

∂i f = f,i

and the action of frame vector fields on functions by

ei ( f ) = eµi ∂µ = f|i .

• For any frame the commutator of the frame vector fields defines the com-
mutation coefficients
[ei , e j ] = C k i j ek .

• The commutation coefficients are scalar functions, in general.

Theorem 6.1.1 A frame ei is a coordinate frame if and only if for any

• i, j = 1, . . . , n,
[ei , e j ] = 0 .
Proof :

1. Need to show that there is a local coordinate system xi such that ei =

∂i .
2. Use the dual basis of 1-forms to prove that they are exact.


Definition 6.1.2 Let {ei } be a frame of vector fields and σ j be the dual
frame of 1-forms. The symbols ωi jk defined by
•
ωi k j = σi (∇e j ek )

are called the coefficients of the affine connection.

• We denote
∇i = ∇ei .

• Then
∇i e j = ωk ji ek ,

diffgeom.tex; April 12, 2006; 17:59; p. 143

6.1. AFFINE CONNECTION 145

• Then, if X = X i ei is a vector field, then

∇X = X i ∇i .

• If Y = Y j e j is another vector field then

n o
∇X Y = X i ei (Y k ) + ωk ji Y j ek
nh i o
= dY k + ωk ji Y j σi (X) ek .

• That is,
∇X Y = X i (∇i Y)k ek

where
(∇i Y)k = ei (Y k ) + ωk ji Y j

• We will often write simply ∇i Y k meaning (∇ei Y)k . This should not be con-
fused with the covariant derivative of the scalar functions Y k .

• The tensor field ∇Y of type (1, 1) with components ∇i Y k , that is,

∇Y = σi ⊗ ∇i Y = ∇i Y k σi ⊗ ek .

is called the covariant derivative of the vector field Y.

• The components of the covariant derivative are also denoted by Y k ;i , in con-

trast to partial derivatives ∂i Y k , which are also denoted by Y k ,i .

• In the coordinate frame ei = ∂i the covariant derivative takes the form

∇i Y k = ∂i Y k + ωk ji Y j .

diffgeom.tex; April 12, 2006; 17:59; p. 144

146 CHAPTER 6. CONNECTION AND CURVATURE

6.1.2 Curvature, Torsion and Levi-Civita Connection

Definition 6.1.3 Let X and Y be vector fields on a manifold M. Then
the vector field

T (X, Y) = ∇X Y − ∇Y X − [X, Y]

defines a tensor field T of type (1, 2), called the torsion, so that for any
• 1-form σ
T (σ, X, Y) = σ(T (X, Y)) .
The affine connection is called torsion-free (or symmetric) if the tor-
sion vanishes, that is, for any X, Y,

∇X Y − ∇Y X = [X, Y]

• The components of the torsion tensor are defined by

T i jk = σi (T (e j , ek )) .

• In the coordinate frame the components of the torsion tensor are given by

T i jk = ωi k j − ωi jk .

Definition 6.1.4 Let X, Y and Z be vector fields on a manifold M.

Then the vector field

R(X, Y)Z = [∇X , ∇Y ] − ∇[X,Y] Z



defines a tensor field R of type (1, 3), called the Riemann curvature, so
• that for any 1-form σ

R(σ, Z, X, Y) = σ(R(X, Y)Z) .

The affine connection is called flat if the curvature vanishes, that is, for
any X, Y, Z,
[∇X , ∇Y ]Z = ∇[X,Y] Z .

• The components of the curvature tensor are defined by

Ri jkl = σi (R(ek , el )e j )) .

diffgeom.tex; April 12, 2006; 17:59; p. 145

6.1. AFFINE CONNECTION 147

Theorem 6.1.2 The components of the curvature tensor have the form
•
Ri jkl = ωi jl|k − ωi jk|l + ωi mk ωm jl − ωi ml ωm jk − C m kl ωi jm .
Proof :

1.


• In the coordinate frame the components of the curvature tensor are given by

Ri jkl = ∂k ωi jl − ∂l ωi jk + ωi mk ωm jl − ωi ml ωm jk .

• For a Riemannian manifold (M, g) the metric tensor g has the components

gi j = g(ei , e j ) = gµν eµi eνj .

• This metric is used to lower and raise the frame indices.

Definition 6.1.5 Let (M, g) be a Riemannian manifold and ∇ be an
affine connection on M. Then the connection ∇ is called compatible
with the metric g if for any vector fields X, Y and Z it satisfies the
• condition
Z(g(X, Y)) = g(∇Z X, Y) + g(X, ∇Z Y) .
An affine connection that is torsion-free and compatible with the metric
is called the Levi-Civita connection.
• Wed define
ωi jk = gim ωm jk Ci jk = gimC m jk .

• The coefficients of the Levi-Civita connection satisfy the equation

gi j|k = ωi jk + ω jik .

Theorem 6.1.3 Then

• 1 
ωi jk = gi j|k + gik| j − g jk|i + Cki j + C jik − Ci jk .
2
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 146

148 CHAPTER 6. CONNECTION AND CURVATURE

1. Direct calculation.


• The coefficients of the Levi-Civita connection in a coordinate frame are

called Christoffel symbols and denoted by Γi jk .

Corollary 6.1.1 The coefficients of the Levi-Civita connection in a

coordinate frame (Christoffel symbols) have the form
1  
• Γi jk = gim ∂ j gmk + ∂k g jm − ∂m g jk .
2
Christoffel symbols have the following symmetry property

Γi jk = Γi k j .
Proof :

1. Direct calculation.

Corollary 6.1.2 The coefficients of the Levi-Civita connection in an
orthonormal frame have the form
1 
• ωi jk = Cki j + C jik − Ci jk .
2
They have the following symmetry properties

ωi jk = −ω jik .
Proof :

1. Direct calculation.

Theorem 6.1.4 Each Riemannian manifold has a unique Levi-Civita
•
connection.
Proof :

1. By construction.


diffgeom.tex; April 12, 2006; 17:59; p. 147

6.1. AFFINE CONNECTION 149

6.1.3 Parallel Transport

• Let x0 and x1 be two points on a manifold M and C be a smooth curve
connecting these points described locally by xi = xi (t), where t ∈ [0, 1] and
x(0) = x0 and x(1) = x1 . The tangent vector to C is defined by

X = ẋ(t),

where the dot denotes the derivative with respect to t.

• Let Y be a vector field on M. We say that Y is parallel transported along C

if
∇X Y = 0 .

• The vector field Y is parallel transported along C if its components satisfy

the linear ordinary differential equation
d i
Y (x(t)) + ωi jk (x(t)) ẋk (t)Y j (x(t)) = 0 .
dt

• Problem. Solve the equation of parallel transport in terms of a Taylor series

up to terms qubic in the connection coefficients.

• A curve C such that the tangent vector to C is trasported parallel along C,

that is,
∇ ẋ ẋ = 0,
is called the geodesics.

• The coordinates of the geodesics x = x(t) satisfy the non-linear second-

order ordinary differential equation

ẍi + ωi jk (x(t)) ẋk ẋ j = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 148

150 CHAPTER 6. CONNECTION AND CURVATURE

6.2 Tensor Analysis

6.2.1 Covariant Derivative of Tensors
• We extend the definition of the affine connection from the tangent bundle
T M to arbitrary tensor bundles T qp M = ⊗ p T M ⊗q T ∗ M.

• The affine connection on a tensor bundle T qp M is an operator

∇ : C ∞ (T M) × C ∞ (T qp M) → C ∞ (T qp M)

that assigns to a vector field X and a tensor field T of type (p, q) a new
tensor field ∇X T of type (p, q).

• The covariant derivative of a tensor field T of type (p, q) is a linear opera-

tor
p
∇ : C ∞ (T qp M) → C ∞ (T q+1 M)
that assigns to a tensor field T of type (p, q) a new tensor field ∇T of type
(p, q + 1).

• First of all, the covariant derivative of a 1-form α on a manifold M is a

tensor ∇α of type (0, 2) such that for any two vector fields X and Y

(∇α)(X, Y) = (∇X α)(Y) = X[α(Y)] − α(∇X Y) .

• Then the covariant derivative of a tensor T of type (p, q) is a tensor ∇T

of type (p, q + 1) such that for any vector fields X, Y1 , . . . , Yq and 1-forms
ω1 , ω p

(∇T )(X, Y1 , . . . , Yq , ω1 , . . . , ω p ) = (∇X T )(X, Y1 , . . . , Yq , ω1 , . . . , ω p )

= X[T (Y1 , . . . , Yq , ω1 , . . . , ω p )]
Xq
− T (Y1 , . . . , ∇X Yi , . . . Yq , ω1 , . . . , ω p )
i=1
p
X
− T (Y1 , . . . , Yq , ω1 , . . . , ∇X ω j , . . . , ω p ) .
j=1

• Let ei be a basis of vector fields and σi be the dual basis of 1-forms.

diffgeom.tex; April 12, 2006; 17:59; p. 149

6.2. TENSOR ANALYSIS 151

• The covariant derivative of a 1-form has the form

(∇i α)k = ei (αk ) − ωl ki αl

In the coordinate frame this simplifies to

(∇i α)k = ∂i αk − ωl ki αl .

• The covariant derivative of a tensor of type (p, q) in a coordinate basis has

the form
p q
j ... j j ... j j1 ... jm−1 l jm+1 ... j p j ... j
X X
∇i T k11...kqp = ∂i T k11...kqp + ω jm
li T k1 ...kq − ωl kn i T k11...kn−1
p
lkn kq
m=1 n=1

• The parallel transport of tensor fields is defined similarly to vector fields.

Let C be a smooth curve on a manifold M described locally by xi = xi (t),
where t ∈ [0, 1], with the tangent vector X = ẋ(t).

• Let T be a tensor field on M. We say that T is parallel transported along

C if
∇X T = 0 .

6.2.2 Ricci Identities

• The commutators of covariant derivatives of tensors are expressed in terms
of the curvature and the torsion.

• In a coordinate basis for a torsion-free connection we have the following

identities (called the Ricci identities):

[∇i , ∇ j ]Y k = Rk li j Y l

[∇i , ∇ j ]αk = −Rl ki j αl

p q
j ... j j ... j l jm+1 ... j p j ... j
X X
[∇i , ∇ j ]T k11...kqp = R jm li j T k11...kqm−1 − Rl kn i j T k11...kn−1
p
lkn kq
m=1 n=1

diffgeom.tex; April 12, 2006; 17:59; p. 150

152 CHAPTER 6. CONNECTION AND CURVATURE

6.2.3 Normal Coordinates

• Let (M, g) be a Riemannian manifold and Γijk be the Christoffel coefficients
defining the Levi-Civita connection in a coordinate basis.
• Let x0 be a point in M and xi be a local coordinate system in a coordinate
patch about x0 .
• We expand the metric in a Taylor series at the point x0
1
gi j (x) = gi j (x0 )+[∂k gi j ](x0 )(xk −x0k )+ [∂k ∂l gi j ](x0 )(xk −x0k )(xl −x0l )+O((x−x0 )3 )
2
• The matrix gi j (x0 ) is a constant real symmetric matrix with real eigenvalues.

Theorem 6.2.1 There exists a coordinate system such that

gi j (x0 ) = δi j , [∂k gi j ](x0 ) = 0 ,

• so that the Taylor series has the form

1
gi j (x) = δi j + [∂k ∂l gi j ](x0 )(xk − x0k )(xl − x0l ) + O((x − x0 )3 ) .
2
Such coordinates are called Riemann normal coordinates.
Proof :
1.

Corollary 6.2.1 In Riemann normal coordinates the Christoffel sym-
bols vanish at x0
Γijk (x0 ) = 0
and the curvature of the Levi-Civita connection at x0 is expressed in
terms of second derivatives of the metric at x0 ,
•
1n o
Ri jkl (x0 ) = ∂k ∂ j gil − ∂l ∂ j gik − ∂i ∂l gk j + ∂k ∂i g jl ,
2 x0

so that the Taylor series of the metric at x0 has the form

1
gi j (x) = δi j − Rik jl (x0 )(xk − x0k )(xl − x0l ) + O((x − x0 )3 ) .
3

diffgeom.tex; April 12, 2006; 17:59; p. 151

6.2. TENSOR ANALYSIS 153

Proof :

1.

6.2.4 Properties of the Curvature Tensor

• Let (M, g) be an n-dimensional Riemannian manifold. We will restrict our-
selves to the Levi-Civita connection below. We define some new curvature
tensors.

• The Ricci tensor

Ri j = Rk ik j .

• The scalar curvature

R = gi j Ri j = gi j Rk ik j .

• The Einstein tensor

1
Gi j = Ri j − gi j R .
2

• The trace-free Ricci tensor

1
E i j = Ri j − gi j R .
n

• The Weyl tensor (for n > 2)

4 2
C i j kl = Ri j kl − R[i [k δ j] l] + Rδ[i [k δ j] l]
n−2 (n − 1)(n − 2)
4 2
= Ri j kl − E [i [k δ j] l] − Rδ[i [k δ j] l]
n−2 n(n − 1)

diffgeom.tex; April 12, 2006; 17:59; p. 152

154 CHAPTER 6. CONNECTION AND CURVATURE

Theorem 6.2.2 The Riemann curvature tensor of the Levi-Civita con-

nection has the following symmetry properties

1. Ri jkl = −Ri jlk

• 2. Ri jkl = −R jikl

3. Ri jkl = Rkli j

4. Ri [ jkl] = Ri jkl + Ri kl j + Ri l jk = 0

5. Ri j = R ji

Proof :

1.


Theorem 6.2.3 The Weyl tensor has the same symmetry properties as
the Riemann tensor and all its contractions vanish, that is,
•
C i jik = 0 .

Proof :

1.


Theorem 6.2.4 The number of algebraically independent components

of the Riemann tensor of the Levi-Civita connection is equal to
•
n2 (n2 − 1)
.
12
Proof :

1.


diffgeom.tex; April 12, 2006; 17:59; p. 153

6.2. TENSOR ANALYSIS 155

Corollary 6.2.2 In dimension n = 2 the Riemann tensor has only one

independent component determined by the scalar curvature
1
R12 12 = R .
2
• The trace-free Ricci tensor Ei j vanishes, that is

Ri j kl = Rδ[i [k δ j] l]

1
Ri j = Rgi j .
2
Proof :

1.

Corollary 6.2.3 In dimension n = 3 the Riemann tensor has six in-
dependent components determined by the Ricci tensor Ri j . The Weyl
• tensor Ci jkl vanishes, that is,

Ri j kl = 4R[i [k δ j] l] + Rδ[i [k δ j] l] .
Proof :

1.


6.2.5 Bianci Identities

• Let (M, g) be an n-dimensional Riemannian manifold. We will restrict our-
selves to the Levi-Civita connection below.
Theorem 6.2.5 The Riemann tensor satisfies the following identities

• ∇[m Ri j kl] = ∇m Ri j kl + ∇k Ri j lm + ∇l Ri j mk = 0 .

These identities are called the Bianci identities.

Proof :

1.

diffgeom.tex; April 12, 2006; 17:59; p. 154

156 CHAPTER 6. CONNECTION AND CURVATURE


Corollary 6.2.4 The divergences of the Riemann tensor and the Ricci
tensor have the form

∇i Ri j kl = ∇k Rlj − ∇l Rkj ,
• 1
∇i Rij = ∇ j R .
2
The divergence of the Einstein tensor vanishes

∇iGij = 0 .
Proof :

1.


• Problem. By using the Bianci identities simplify the Laplacian of the Rie-
mann tensor, ∆Ri j kl = ∇m ∇m Ri j kl .

diffgeom.tex; April 12, 2006; 17:59; p. 155

6.3. CARTAN’S STRUCTURAL EQUATIONS 157

6.3 Cartan’s Structural Equations

• Let ∂µ be a coordinate basis for the tangent bundle T M and ei = eµi ∂µ be an
orthonormal frame of vector fields.
• Then
gi j = g(ei , ek ) = gµν eµi eνj = δi j .

• We use this metric to lower and raise the frame indices.

• Let dxµ be a coordinate basis for the cotangent bundle T M and σi = σiµ dxµ
be an orthonormal frame of 1-forms dual to ei .
• Then
σi (e j ) = σiµ eµj = δij
and
gµν σiµ σνj = δi j
σiµ eνi = δνµ .

• The commutators of the frame vector fields define the commutation coeffi-
cients C i jk by
[ei , e j ] = C k i j ek ,
or, in components,

eνj|i − eνi| j = eµi ∂µ eνj − eµj ∂µ eνi = C k i j eνk .

That is,
C k i j = σk ([ei , e j ])
or
C k i j = σkν eµi ∂µ eνj − eµj ∂µ eνi .
 

Proposition 6.3.1 There holds

1
dσi = − C i jk σ j ∧ σk .
• 2
or
1
(dσi )(e j , ek ) = − C i jk .
2
Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 156

158 CHAPTER 6. CONNECTION AND CURVATURE

1. Direct calculation using the duality condition.



• Let ωi jk be the coefficients of the Levi-Civita connection in the orthonormal

frame.

• Then
∇i e j = ωk ji ek
and
∇i σ j = −ω j ki σk ,
where ∇i = ∇ei .

• This can also be written as

ωk ji = σkµ eνi eµj;ν = −σkµ;ν eνi eµj .

• Thus

σkν;µ − σkµ;ν eνi eµj

 
ωkji − ωk i j =
= σkν,µ − σkµ,ν eνi eµj
 

= (dσk )(e j , ei ) .

Proposition 6.3.2 There holds

•
dσi = ωi jk σ j ∧ σk .
Proof :

1. Follows from above.



• Since the torsion of the Levi-Civita connection is zero, we have

∇i e j − ∇ j ei = [ei , e j ] .

• Therefore,
ωk ji − ωk i j = C k i j .

diffgeom.tex; April 12, 2006; 17:59; p. 157

6.3. CARTAN’S STRUCTURAL EQUATIONS 159

• Since the Levi-Civita connection is compatible with the metric, we have

0 = ∇i g(e j , ek ) = g(∇i e j , ek ) + g(e j , ∇i ek ) .

• Thus,
ωk ji + ω jki = 0 .

• Finally, we obtain,

Proposition 6.3.3 The coefficients of the Levi-Civita connection in an

orthonormal frame are given in terms of the commutation coefficients
• by
1 
ωi jk = Cki j + C jik − Ci jk .
2
Proof :
1. Use the equations

ωki j + ωik j = 0
ωi jk + ω jik = 0
ω jki + ωk ji = 0

and
ωk ji − ωk i j = C k i j .

• Now we define the connection 1-forms

Ai j = ωi jk σk

and the curvature 2-forms

1
F i j = Ri jkl σk ∧ σl .
2

• Then the equation

dσi = ωi jk σ j ∧ σk
can be written as
dσi + Ai j ∧ σ j = 0 .
This is called Cartan’s first structural equation.

diffgeom.tex; April 12, 2006; 17:59; p. 158

160 CHAPTER 6. CONNECTION AND CURVATURE

• The curvature 2-forms are obtained from the connection 1-forms by Car-
tan’s second structural equation

F i j = dAi j + Ai k ∧ Ak j .

This equation is equivalent to the expression for the curvature components

and can be obtained from that.

• Cartan’s third structural equation

dF i j + Ai k ∧ F k j − F i k ∧ Ak j = 0 .

is equivalent to Bianci identities.

• Cartan structural equations can be written in a very compact way by in-

troducing the covariant exterior derivative acting on vector valued and
matrix valued forms.

• Let α be a p-form valued in a vector space V (in our case V = Rn ). Such

a form is called a twisted form. Let αi be the components of this form in
a fixed basis in V. That is, αi is a p-form for each i = 1, . . . , n. Then the
covariant exterior derivative

D : Λ p ⊗ V → Λ p+1 ⊗ V

is defined by
(Dα)i = dαi + Ai j ∧ α j
or, in matrix form,
Dα = dα + A ∧ α
with obvious notation.

• Let V ∗ be the dual vector space to V (in our case it is again Rn ). We consider
p-forms valued in V ∗ (covectors) and naturally extend the operator D to
such forms
D : Λ p ⊗ V ∗ → Λ p+1 ⊗ V ∗
by
(Dα)i = dαi − (−1) p α j ∧ A j i
or, in matrix form,
Dα = dα − (−1) p α ∧ A .

diffgeom.tex; April 12, 2006; 17:59; p. 159

6.3. CARTAN’S STRUCTURAL EQUATIONS 161

• Finally, we consider matrix-valued p-forms valued in V ⊗ V ∗ and extend the

operator D to such forms

D : Λ p ⊗ V ⊗ V ∗ → Λ p+1 ⊗ V ⊗ V ∗

by
(Dα)i j = dαi j + Ai k ∧ αkj − (−1) p αik ∧ Ak j
or, in matrix form,

Dα = dα + A ∧ α − (−1) p α ∧ A .

• Now, let σ = (σi ), F = (F i j ) and α = (αi ) be an arbitrary vector-valued

1-form. Then

Dσ = 0
D2 α = F ∧ α
DF = 0 .

• Problem. Let the dimension n = 2k of the manifold M be even. We define

the following 2l-forms

Ω(l) = tr F
| ∧ {z
· · · ∧ F}
l

and the n-form

Ω = εi1 ...i2k Fi1 i2 ∧ · · · ∧ Fi2k−1 i2k

1. Prove that these forms are independent of the orthonormal basis and
are closed, that is,
dΩ(l) = dΩ = 0 .
2. Find the expressions in local coordinates for these forms.

• These forms define so called characteristic classes, which are closed in-
variant forms whose integrals over the manifold do not depend on the metric
and, therefore, are topological invariants of the manifold.

diffgeom.tex; April 12, 2006; 17:59; p. 160

162 CHAPTER 6. CONNECTION AND CURVATURE

diffgeom.tex; April 12, 2006; 17:59; p. 161

Chapter 7

Homology Theory

7.1 Algebraic Preliminaries

7.1.1 Groups
• A group is a set G with a binary operation, ·, called the group multiplication,
that is,
1. associative,
2. has an identity element,
3. every element has an inverse.
• A group is Abelian if the group operation is commutative.
• For an Abelian group the group operation is called addition and is denoted
by +. The identity element is called zero and denoted by 0. The inverse
element of an element g ∈ G is denoted by (−g).
• Let G and E be Abelian groups. A map F : G → E is called a homomor-
phism if for any g, g0 ∈ G,

F(g +G g0 ) = F(g) +E F(g0 ) ,

where +G and +E are the group operations in G and E respectively.

• In particular,

F(0G ) = 0E , and F(−g) = −F(g) .

163
164 CHAPTER 7. HOMOLOGY THEORY

• Let F : G → E be a homomorphism of an Abelian group G into an Abelian

group E. The set of elements of G mapped to the identity element of E is
denoted by
Ker F = {g ∈ G | F(g) = 0E } ,
where 0E is the identity element of E, and called the kernel of the homo-
morphism F.

• The image of the homomorphism F : G → E is the set

Im F = {h ∈ E | h = F(g) for some g ∈ G} .

• A homomorphism F : G → E of a group G into a group E is called an

isomorphism if it is a bijection.

• A subset H of a group G is called a subgroup of G if it contains the identity

element and is closed under the group operation.

• For any homomorphism F : G → E the kernel Ker F is a subgroup of G.

• Let G be an Abelian group and H be a subgroup of G. We say that two

elements of G are equivalent if they differ by an element of H.

• Let g ∈ G be an element of G. Then the set of all elements of G equivalent

to g, denoted by [g] = g + H, is an equivalence class of g called a coset.

• The set of cosets is denoted by G/H.

• An element g used to describe a coset [g] is called a representative.

• The set of cosets G/H is an Abelian group, called the quotient group, with
the addition defined by

[g] + [g0 ] = [g + g0 ] .

• The projection map π : G → G/H defined by

π(g) = [g]

is a homomorphism.

diffgeom.tex; April 12, 2006; 17:59; p. 162

7.1. ALGEBRAIC PRELIMINARIES 165

Theorem 7.1.1 Fundamental Theorem of Homomorphisms. Let G

• and F be groups. Let F : G → F be a homomorphism. Then

G/Ker F  Im F .
Proof :

1. Since
F(g + Ker F) = F(g) .

Theorem 7.1.2 Let G and E be Abelian groups, and H ⊂ G and N ⊂ E
be their subgroups so that G/H and E/N are the quotient groups. Let
F : G → E be a homomorphism such that the image of the subgroup H
of G is the subgroup N of E, that is,
•
F(H) = N .

Then the homomorphism F induces a homomorphism of the quotient

groups
F∗ : G/H → E/N .
Proof :

1. Let
π : E → E/N
be the projection homomorphism defined by, for any x ∈ E

π(x) = [x] = x + N .

2. Then
F∗ = π ◦ F : G/H → E/N
is a homomorphism.


• A field is a set K with two binary operations, addition, +, and multiplication,

·, that satisfy the following conditions:

1. both addition and multiplication are associative,

2. both addition and multiplication are commutative,

diffgeom.tex; April 12, 2006; 17:59; p. 163

166 CHAPTER 7. HOMOLOGY THEORY

3. both addition and multiplication have identity elements, the additive

identity 0 and the multiplicative identity 1, such that 0 , 1,
4. the multiplication is distributive with respect to addition,
5. every element has an additive inverse,
6. every nonzero element has a multiplicative inverse.

• In particular, any field is an Abelian group with respect to addition.

• A vector space over a field K consists of a set E, whose elements are called
vectors, and the field K, whose elements are called scalars, with two oper-
ations: vector addition, + : E × E → E, and multiplication by scalars,
· : K × E → E, that satisfy the following conditions:

1. the vector addition is associative and commutative,

2. there is an additive identity, called the zero vector,
3. every vector has an additive inverse, called the opposite vector,
4. for any v ∈ E, a, b ∈ K,
(a) a(bv) = (ab)v,
(b) (a + b)v = av + bv,
(c) a(u + v) = au + av,
(d) 1v = v.

• Let E be a vector space and F be a vector subspace of E. Then E/F is a

vector space.

• Let E be an inner product vector space and F be its subspace. Then

E/F = F ⊥

is the orthogonal complement of F in E, and

π : E → F⊥

is the orthogonal projection to F ⊥ .

diffgeom.tex; April 12, 2006; 17:59; p. 164

7.1. ALGEBRAIC PRELIMINARIES 167

7.1.2 Finitely Generated and Free Abelian Groups

• Let G be an Abelian group. Let g1 , . . . , gr ∈ G be some elements of G and
 r 

X 
H=
 
n g | g ∈ G, n ∈

 k k k k Z

 
k=1

be a set of linear combinations of gk .

• Then H is a subgroup of G. The elements gk are called the generators of H
and H is said to be generated by gk .
• If a group G is generated by finitely many elements of G, then G is called a
finitely generated group.
• The elements g1 , . . . , gr are linearly independent if for any integer coeffi-
cients n1 , . . . , nr the linear combination
Xr
nk gk , 0
k=1

is not equal to zero.

• A finitely generated group G is called a free Abelian group of rank r if it
is generated by r linearly independent elements.

7.1.3 Cyclic Groups

• An Abelian group generated by one element is called a cyclic group.
• Infinite cyclic groups.
• Finite cyclic group.
Theorem 7.1.3 Fundamental Theorem of Finitely Generated
Abelian Groups. Let G be a finitely generated Abelian group with m
generators. Then G is isomorphic to the direct sum of cyclic groups,
•
| ⊕ {z
GZ Z ⊕Zk1 ⊕ · · · ⊕ Zk p ,
··· ⊕}
r

where m = r + p. The number r is called the rank of G.

Proof :

diffgeom.tex; April 12, 2006; 17:59; p. 165

168 CHAPTER 7. HOMOLOGY THEORY

1.


diffgeom.tex; April 12, 2006; 17:59; p. 166

7.2. SINGULAR CHAINS 169

7.2 Singular Chains

• The standard Euclidean p-simplex in R p is the convex set ∆ p ⊂ R p gener-
ated by an ordered (p + 1)-tuple (P0 , P1 , . . . , P p ) of points in R p

P0 = (0, . . . , 0), Pi = (0, . . . , 0, 1, 0, . . . , 0), i = 1, . . . , p ,

all of whose components are 0 except for the i-th component, which is equal
to 1.

• We use the following notation

∆ p = (P0 , P1 , . . . , P p ) .

• Let M be an n-dimensional manifold. A singular p-simplex in M is a

differentiable map
σp : ∆p → M .

• By slightly abusing notation we denote the image σ p (∆ p ) of ∆ p in M under

the map σ p just by σ p .

• Let α be a p-form on M and σ p be a p-simplex in M. We define the integral

of α over σ p by Z Z
α= σ∗p α .
σp ∆p

• The k-th face of a standard p-simplex ∆ p = (P0 , P1 , . . . , P p ) (a face opposite

to to the vertex Pk ) is the convex set in R p generated by an ordered p-tuple
of points in R p
∆(k)
p−1 = (P0 , . . . , P̂k , . . . , P p )

where the k-th point Pk is omitted.

• Since this points lie in R p and R p−1 a face is not a standard (p − 1)-simplex.
It can be rather described as a singular simplex in R p defined by the unique
affine map
fk : ∆ p−1 → ∆ p
whose image in R p is exactly ∆(k)
p .

diffgeom.tex; April 12, 2006; 17:59; p. 167

170 CHAPTER 7. HOMOLOGY THEORY

• Such a map is uniquely defined as follows. Let ∆ p−1 = (Q0 , . . . , Q p−1 ),

where Qi ∈ R p−1 , be the standard (p−1)-simplex in R p−1 and ∆ p = (P0 , . . . , P p ),
where Pi ∈ R p , be the standard p-simplex in R p . Then
fk (Qi ) = Pi for i = 0, . . . , k − 1
and
fk (Qi ) = Pi+1 for i = k, . . . , p − 1 .
• If Q is a point in R p−1 with coordinates (xi ) = (x1 , . . . , x p−1 ), then P = fk (Q)
is the point in R p with coordinates (yµ ) = (y1 , . . . , y p ), where
yµ = Aµ(k) j x j + bµ(k) .
The above requirements fix the matrix A(k) and the vector b(k) uniquely.
• Problem. Find the maps fk .
• Let M and V be manifolds and ϕ : V → M be a differentiable map. Let
σ p : ∆ p → V be a singular p-simplex in V.
• Then the composition map
ϕ ◦ σp : ∆p → M
defines a singular p-simplex in M.
• Thus, the composition
σ p ◦ fk : ∆ p−1 → M
defines a singular (p−1)-simplex in M, which is the k-th face of the singular
p-simplex σ p in M.
• The boundary ∂∆ p of the standard p-simplex ∆ p is defined as follows. For
p > 0, we define
p
X
∂(P0 , P1 , . . . , P p ) = (−1)k (P0 , . . . , P̂k , . . . , P p ) ,
k=0

that is, the boundary is the formal sum

Xp
∂∆ p = (−1)k ∆(k)
p−1 .
k=0

For p = 0, we define
∂∆0 = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 168

7.2. SINGULAR CHAINS 171

• Examples.

• The boundary of a standard simplex is not a simplex, but an integer (p − 1)-

chain.

• Let G be an Abelian group. A singular p-chain on M with coefficients in

G is a finite formal sum
Xr
cp = gk σkp
k=1

of singular simplexes σkp : ∆ p → M with coefficients gk which are elements

of the group G.

• Examples.

• Let S p (M) be the set of all singular p-simplexes in M. Then, a p-chain is a

function
c p : S p (M) → G ,
such that its value is not equal to zero only for finitely many simplexes.
These simplexes are exactly σkp listed in the formal sum, and the values of
the function c p are exactly the coefficients of the formal sum, that is

c p (σkp ) = gk .

• Alternatively, a p-chain can be thought of as a finite subset of S p (M) × G,

that is, a finite set of ordered pairs

c p = {(σkp , gk )}rk=1

• The notation of a p-chain as a sum, or as a function, is useful since we can

define the addition of p-chains simply as addition of corresponding func-
tions. If two p-chains c p and c0p have the same p-simplex σ p in both of
them, then in the sum c p + c0p we add the corresponding group elements g
and g0 . That is, if

c p = gσ p + . . . , and c0p = g0 σ p + . . . ,

then
c p + c0p = (g + g0 )σ p + . . . .

diffgeom.tex; April 12, 2006; 17:59; p. 169

172 CHAPTER 7. HOMOLOGY THEORY

• We denote the identity element of G by 0 and define the zero p-chain simply
by
0p = 0 .

• Then the set of all singular p-chains on M with coefficients in G forms an

Abelian group, called the singular p-chain group of M with coefficients in
G and denoted by C p (M; G).

• A chain with integer coefficients, when G = Z, is called an integer chain.

• The standard p-simplex ∆ p is an integer p-chain in R p with only one term:

c p = 1 · ∆ p . That is, it is an element of C p (R p ; Z).

• The boundary of the standard p-simplex in R p ,

p
X
∂∆ p = (−1)k ∆(k)
p−1
k=0

is an integer (p − 1)-chain in R p , that is, an element of C p−1 (R p ; Z).

• Let M and V be closed manifolds. Let F : M → V be a map of M into

V and σ p : ∆ p → M be a singular p-simplex in M. Then the composition
F ◦ σ p : ∆ p → V is a singular p-simplex in V. We denote it by

F∗ σ p = F ◦ σ p .

• The induced chain homomorphism

F∗ : C p (M; G) → C p (V; G)

is defined by: for any gk ∈ G and σkp ∈ S p (M),

 r  r
X  X
F∗  gk σ p  =
 k  gk F∗ σkp
k=1 k=1

• Let F : M → V and E : V → W be two maps of manifolds and F∗ :

C p (M; G) → C p (V; G) and E∗ : C p (V; G) → C p (W; G) be the correspond-
ing induced chain homomorphisms. Then

(E ◦ F)∗ = E∗ ◦ F∗ .

diffgeom.tex; April 12, 2006; 17:59; p. 170

7.2. SINGULAR CHAINS 173

• Let σ p : ∆ p → M be a singular p-simplex in M. Then its boundary ∂σ p is

the integer (p − 1)-chain in M defined by
∂σ p = σ p∗ (∂∆ p ) .

• In more detail
∂σ p = σ p∗ (∂∆ p )
Xp p
X
= (−1) σ p∗ (∆ p−1 ) =
k (k)
(−1)k (σ p ◦ fk ) .
k=0 k=0

p−1 ) = (σ p ◦ fk ) is the k-th face of the singular p-simplex

Recall that σ p∗ (∆(k)
σp.
• That is, the boundary of the image of ∆ p is the image of the boundary of ∆ p .
• The boundary of any singular p-chain with coefficients in G is defined by,
for any gk ∈ G, σkp ∈ S p (M),
 r  r
X  X
∂  gk σ p  =
k
gk ∂σkp .
k=1 k=1

• This leads to the boundary homomorphism

∂ : C p (M; G) → C p−1 (M; G) .

• Let F : M → V be a map, σ p be a singular p-simplex in M and F∗ σ p be the

induced singular p-simplex in V. Then
∂(F∗ σ p ) = ∂(F ◦ σ p )
= (F ◦ σ p )∗ (∂∆ p )
= (F∗ ◦ σ p∗ )(∂∆ p )
= F∗ [σ p∗ (∂∆ p )]
= F∗ (∂σ p )

• More generally, let

r
X
cp = gk σkp
k=1
be a p-chain on M and F∗ c p be the induced p-chain on V.

diffgeom.tex; April 12, 2006; 17:59; p. 171

174 CHAPTER 7. HOMOLOGY THEORY

• Then
∂(F∗ c p ) = F∗ (∂c p ) .

• Therefore,
∂ ◦ F∗ = F∗ ◦ ∂ ,
in other words, the boundary of an image is the image of the boundary.

• Thus, we obtain a commutative diagram

F∗
C p (M; G) →
C p (V; G)
∂↓ ∂↓
C p−1 (M; G) → C p−1 (V; G)
F∗

which is a fancy way to say that for any p-chain c p ∈ C p (M; G) on M we

have
F∗ (∂c p ) = ∂(F∗ c p ) .

Theorem 7.2.1 The boundary of a boundary is zero, that is,

•
∂2 = 0 .
Proof :

1. For a standrad p-simplex ∆ p we have

p
X
∂∂∆ p = (−1)k ∂∆(k)
p
k=0
Xp
= (−1)k ∂(P0 , . . . , P̂k , . . . , P p )
k=0
Xp k−1
X
= (−1) k
∂(P0 , . . . , P̂ j , . . . , P̂k , . . . , P p )
k=0 j=0
Xp X p
+ (−1)k ∂(P0 , . . . , P̂k , . . . , P̂ j , . . . , P p )
k=0 j=k+1
= 0

because of the pairwise cancellation.

diffgeom.tex; April 12, 2006; 17:59; p. 172

7.2. SINGULAR CHAINS 175

2. Then, for a singular p-simplex σ p

∂∂σ p = ∂[σ p∗ (∂∆ p )] = σ p∗ ∂(∂∆ p ) = σ∗ (0) = 0

7.2.1 Examples
• Cylinder.

• Möbius Band.

diffgeom.tex; April 12, 2006; 17:59; p. 173

176 CHAPTER 7. HOMOLOGY THEORY

7.3 Singular Homology Groups

7.3.1 Cycles, Boundaries and Homology Groups
• We can define the singular p-chains with coefficients in a field K.

• Furthermore, we can define the multiplication of p-chains by elements of

the field K, called the scalars by, for any a, bi ∈ K,
 r  r
X  X
a  bi σ p  =
i
abi σip .
i=1 i=1

• The chain groups C p (M, K) with coefficients in a field K become infinite-

dimensional vector spaces.

• In this case the boundary homomorphism becomes a linear transformation

(operator) in a vector space.

• Let M be a manifold and G an Abelian group. A singular p-chain z p in M

whose boundary is 0 is called a singular p-cycle.

• The set of all p-cycles in M

Z p (M; G) = {z p ∈ C p (M; G) | ∂z p = 0}

is a subgroup of the chain group C p (M; G) called the p-cycle group.

• Obviously the p-cycle group is the kernel of the boundary homomorphism

Z p (M; G) = Ker ∂ p ,

where
∂ p : C p (M; G) → C p−1 (M; G) .

• In the case, when G = K is a field, then Z p (M; K) is a vector subspace of

the vector space C p (M; K).

• A singular p-chain b p in M that is the boundary of a singular (p + 1)-chain

is called a p-boundary.

diffgeom.tex; April 12, 2006; 17:59; p. 174

7.3. SINGULAR HOMOLOGY GROUPS 177

• The set of all p-boundaries in M

B p (M; G) = {b p ∈ C p (M; G) | b p = ∂c p+1 for some c p+1 ∈ C p+1 (M; G)}

is a subgroup of the chain group C p (M; G) called the p-boundary group.

• Obviously the p-boundary group is the image of the boundary homomor-

phism
B p (M; G) = Im ∂ p+1 .

• Since every p-boundary is a p-cycle (because of ∂2 = 0) the group B p (M; G)

is a subgroup of Z p (M; G).

• In the case, when G = K is a field, then B p (M; K) is a vector subspace of

the vector space Z p (M; K).

• Let M be a manifold and G be an Abelain group. We say that two p-cycles

are homologous if they differ by a boundary.

• The set of equivalence classes of p-cycles homologous to each other, that

is, the quotient group

H p (M; G) = Z p (M; G)/B p (M; G) ,

is called the p-th homology group.

• In the case when the coefficient group G is a field G = K, all the groups, Z p ,
B p and H p are vector spaces.

7.3.2 Simplicial Homology

• The important fact is that if M is a compact manifold, then the vector space
H p (M; K) is finite dimensional. (This can be proved but we will not do that).

• Let M be a compact n-dimensional manifold. Then there is a triangulation

of M by finitely many n-simplexes diffeomorphic to the standard n-simplex
∆n .

• Thus, M is a union of finitely many n-simplexes, which are either disjoint

of intersect along common r-simplexes with r = 0, 1, . . . , n − 1.

diffgeom.tex; April 12, 2006; 17:59; p. 175

178 CHAPTER 7. HOMOLOGY THEORY

• The set of all these simplexes form a finite simplicial complex with some
coefficient group G.

• All the simplicial chain groups C̄ p (M; G), Z̄ p (M; G) and B̄ p (M; G) are finitely
generated Abelian groups.

• Therefore, the homology group H̄ p (M; G) is a finitely generated group.

• Since any simplicial cycle can be described as a singular cycle, there are
homomorphisms
Z̄ p → Z p , B̄ p → B p
and the induced homomorphism

H̄ p → H p .

Theorem 7.3.1 Let M be a compact manifold and G be an Abelian

group. Then the singular homology groups are isomorphic to the sim-
plicial homology groups
•
H̄ p (M; G) = H p (M; G)

and are finitely generated Abelian groups.

Proof : Nontrivial.

Corollary 7.3.1 Let M be a compact manifold and K be field. Then
•
the homology groups H p (M; K) are finite-dimensional vector spaces.
Proof : Follows from above theorem.


7.3.3 Betti Numbers and Topological Invariants

• Let M be a compact manifold. The dimensions of the real homology groups
H p (M; R) are called the Betti numbers, that is,

B p (M) = dim H p (M; R) .

• The Betti number B p is the maximal number of linearly independent p-

cycles modulo a boundary.

diffgeom.tex; April 12, 2006; 17:59; p. 176

7.3. SINGULAR HOMOLOGY GROUPS 179

• Let M and V be manifolds, G be a n Abelain group, F : M → V be a map

and F∗ : C p (M; G) → C p (V; G) be the induced homomorphism of chain
groups.

• Since the induced homomorphism F∗ commutes with the boundary homo-

morphism ∂, the groups Z p (M; G) and B p (M; G) are closed under F∗ .

• Therefore, the homomorphism F∗ naturally acts on the homology groups

F∗ : H p (M; G) → H p (V; G) .

• If F : M → V is a homeomorphism, then there is the inverse homeomor-

phism F −1 : V → M and the inverse induced homomorphism

F∗−1 : H p (V; G) → H p (M; G) .

• In this case, the induced homomorphism F∗ is an isomorphism.

Theorem 7.3.2 Let M and V be compact homeomorphic manifolds

and G be an Abelian group. Then their homology groups are isomor-
• phic, that is, for any p

H p (M; K)  H p (V; G) .
Proof : Follows from above.


• Thus, homology groups are topological invariants.

Corollary 7.3.2 Let M and V be compact manifolds. If there is an

Abelian group G and an integer p such that their homology groups are
not isomorphic, that is,
•
H p (M; K)  H p (V; G) ,

then the manifolds M and V are not homeomorphic to each other.

• Remark. The converse is not true.

diffgeom.tex; April 12, 2006; 17:59; p. 177

180 CHAPTER 7. HOMOLOGY THEORY

7.3.4 Some Theorems from Algebraic Topology

• A manifold M is path-connected (or just connected) if every two points in
M can be connected by a piecewise-smooth curve.

• Let M be a manifold and G be an Abelian group.

• A point p in a manifold M is a 0-chain. Since ∂p = 0, then each point is a

0-cycle (by definition).

• A smooth map C : [0, 1] → M defines a singular 1-simplex and

∂C = C(1) − C(0)

is a 0-chain.

• More generally, for any g ∈ G, then gC is a singular 1-simplex and

∂(gC) = gC(1) − gC(0) .

• A piecewise-smooth map C : [0, 1] → M defines a 1-chain (as a formal sum

r
X
c1 = gCk
k=1

of smooth pieces with the same coefficient) and

∂c1 = gC(1) − gC(0)

is a 0-chain.
Theorem 7.3.3 Let M be a compact connected manifold and G be an
Abelian group. Then
•
H0 (M; G) = Gp = {gp | g ∈ G, p ∈ M} .
Proof :

1. In a connected manifold any two 0-simplexes with the same coefficient

are homologous.
2. Moreover, for any point p ∈ M, g , 0, and any element g ∈ G, there
is no 1-chain c1 such that ∂c1 = gp.

diffgeom.tex; April 12, 2006; 17:59; p. 178

7.3. SINGULAR HOMOLOGY GROUPS 181

3. Therefore, a multiple gp of a single point is not a boundary for any

g , 0.
4. Thus, any point p ∈ M is a 0-cycle that is not a boundary.
5. Moreover, for any g ∈ G and any p ∈ M the 0-chain gp is a 0-cycle
that is not a boundary.


• In particular,
H0 (M; Z) = {0, ±p, ±2p, . . . }
and
H0 (M; R) = R
is a one-dimensional vector space.

Corollary 7.3.3 Let M be a compact connected manifold. Then the

zero Betti number is equal to
•
B0 (M) = 1 .
Proof : Follows from above.


Theorem 7.3.4 Let M be a compact manifold consisting of k con-

nected pieces M1 , . . . Mk . Then

H0 (M; R) = Rp1 + · · · + Rpk ,

•
where pi ∈ Mi , i = 1, . . . , k, meaning
 k 
 X 
Rp1 + · · · + Rpk =  , = . . . , .
 
a p a ∈ p ∈ M i 1, k
 
 i i i R, i i 

 
i=1

Proof :


• In this case
H0 (M; R) = Rk
is a k-dimensional vector space.

diffgeom.tex; April 12, 2006; 17:59; p. 179

182 CHAPTER 7. HOMOLOGY THEORY

Corollary 7.3.4 Let M be a compact manifold consisting of k con-

nected pieces. Then the zero Betti number is equal to
•
B0 (M) = k .
Proof : Follows from above.


• Let V be a p-dimensional oriented closed (compact without boundary) man-

ifold.

• Then a triangulation of V defines an integer p-cycle, denoted by [V] so that

∂[V] = 0 .

• This does not work for non-orientable closed manifolds. A triangulation of

a non-orientable closed manifold V gives an integer p-chain, which is not a
p-cycle since
∂[V] , 0 .

• Example. Klein bottle.

• By changing the coefficient group G sometimes one can get a p-cycle even
for non-orientable manifolds.

Theorem 7.3.5 Let M be an n-dimensional compact manifold and V

be a closed oriented p-dimensional submanifold of M. Let G be an
• Abelian group and g be an element of G. Consider a triangulation of V
in p-triangles and assign to each p-triangle in this triangulation of V
the same coefficient g ∈ G. Then g[V] defines a p-cycle z p in H p (M; G).
Proof : Without proof.


• Remark. A p-cycle is a generalization of the concept of a closed oriented

submanifold.

• In the case of real homology groups, when G = R, the following theorem is

true.

diffgeom.tex; April 12, 2006; 17:59; p. 180

7.3. SINGULAR HOMOLOGY GROUPS 183

Theorem 7.3.6 Let M be an n-dimensional compact manifold. Every

real p-cycle z p in H p (M; R) is homologous to a finite formal sum
r
X
• zp ∼ ak Vk
k=1

of closed oriented p-dimensional submanifolds Vk of M with real coef-

ficients ak .
Proof : Nontrivial.


Theorem 7.3.7 Let M be an n-dimensional manifold and G be an

Abelian group. Let z p and z0p be two cycles in H p (M; G) that can be
• deformed into each other. Then they are homologous to each other

z p ∼ z0p .

Proof :

1. Since the deformation defines a deformation chain c p+1 such that

∂c p+1 = z0p − z p .

Proposition 7.3.1 Let M be an n-dimensional closed manifold and G

be an Abelian group. Then for p > n the singular homology groups
•
H p (M; G) are trivial
H p (M; G) = 0 .

Proof :

1. Singular homology groups are isomorphic to the simplicial homology

groups.
2. Since there are no simplicial complexes of dimension p > n then all
simplicial homology groups are trivial for p > n.


diffgeom.tex; April 12, 2006; 17:59; p. 181

184 CHAPTER 7. HOMOLOGY THEORY

7.3.5 Examples.
• Sphere S n .
• From the facts that S n is connected, orientable and closed it follows that
H0 (S n ; G) = Hn (S n ; G) = G ,
H p (S n ; G) = 0 , for p , 0, n ,
B0 (S ) = Bn (S ) = 1 ,
n n

B p (S n ) = 0 , for p , 0, n .

• Torus T 2 .
H0 (T 2 ; G) = H2 (T 2 ; G) = G ,
H1 (T 2 ; G) = GA + GB ,
B0 (T 2 ) = 1, B1 (T 2 ) = 2, B2 (T 2 ) = 1,
where A and B are the basic 1-cycles.
• Klein Bottle K 2 .
• Since K 2 is connected closed non-orientable it follows that
H0 (K 2 ; Z) = Z,
H2 (K 2 , Z) = 0 ,
H1 (K 2 ; Z) = ZA + Z2 B ,
where A and B are the basic 1-cycles, and
H0 (K 2 ; R) = R,
H2 (K 2 , R) = 0 ,
H1 (K 2 ; R) = RA ,
B0 (K 2 ) = 1, B1 (K 2 ) = 1, B2 (K 2 ) = 0 .

• Real Projective Plane RP2 .

• RP2 is connected closed non-orientable.
H0 (RP2 ; Z) = Z,
H2 (RP2 , Z) = 0 ,
H1 (RP2 ; Z) = Z2 A ,

diffgeom.tex; April 12, 2006; 17:59; p. 182

7.3. SINGULAR HOMOLOGY GROUPS 185

where A is the basic 1-cycle, and

H0 (RP2 ; R) = R,
H2 (RP2 , R) = 0 ,
H1 (RP2 ; R) = 0 ,
B0 (RP2 ) = 1, B1 (RP2 ) = 0, B2 (RP2 ) = 0 .

• Torus T 3 .

• T 3 is a connected closed orientable manifold.

H0 (T 3 ; Z) = H3 (T 3 , Z) = Z ,
H1 (T 3 ; Z) = ZA + ZB + ZC ,
H2 (T 3 ; Z) = ZD + ZE + ZF ,

where A, B and C are basic 1-cycles, and D, E and F are basic 2-cyles,

B0 (T 3 ) = 1, B1 (T 3 ) = B2 (T 3 ) = 3, B2 (T 3 ) = 1 .

• Real Projective Space RP3 .

• RP3 is connected closed orientable.

H0 (RP3 ; R) = H3 (RP3 , R) = R ,
H1 (RP3 ; R) = H2 (RP3 ; R) = 0 ,
B0 (RP3 ) = B3 (RP3 ) = 1, B1 (RP3 ) = B2 (RP3 ) = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 183

186 CHAPTER 7. HOMOLOGY THEORY

7.4 de Rham Cohomology Groups

• Let M be a manifold and G = R be the coefficient group.

• Then C p (M; R), Z p (M; R), B p (M; R) and H p (M; R) are vector spaces.

• For simplicity we will denote them in this section simply by C p (M), Z p (M),
B p (M) and H p (M).

• Let C p (M) = C ∞ (Λ p (M)) be the space of smooth p-forms on M.

• We will call the closed p-form p-cocyles and the space

Z p (M) = {α p ∈ C p (M) | dα p = 0}

of all closed p-forms on M, the cocycle group.

• The exact p-forms on M are called the p-coboundaries and the space

B p (M) = {α p ∈ Z p (M) | α p = dβ p+1 for some β p+1 ∈ C p+1 (M)}

of all exact p-forms on M is called the coboundary group.

• Both Z p (M) and B p (M) are vector spaces with real coefficients.

• Recall that the exterior derivative is a map

d p : C p (M) → C p+1 (M)

such that
Ker d p = Z p (M)
and
Im d p−1 = B p (M) .

• Two closed forms are said to be equivalent (or cohomologous) if they differ
by an exact form.

• The collection of all equivalence classes of closed forms is the quotient

vector space
H p (M) = Z p (M)/B p (M)
called the p-th de Rham cohomology group.

diffgeom.tex; April 12, 2006; 17:59; p. 184

7.4. DE RHAM COHOMOLOGY GROUPS 187

• de Rham cohomology groups are vector spaces.

• Let r
X
cp = ak σkp
k=1

be a real p-chain in M, and α be a p-form on M.

• We define the integral of α over c p by

D E Z r
X Z
α, c p = α= ak α.
cp k=1 σkp

• Thus every p-form on M defines a linear functional on C p (M)

α : C p (M) → R ,

by
α D E
c p 7−→ α, c p .

• The space of all p-forms can be naturally identified with the dual space
C ∗p (M)
C p (M)  C ∗p (M) .

• Furthermore, by Stokes theorem we have for a (p − 1)-form

D E D E
dα, c p = α, ∂c p .

• Thus, for every p-cycle z p , that is, if ∂z p = 0,

D E
dα, z p = 0 ,

and for every closed form α, that is, if dα = 0,

D E
α, ∂c p = 0 .

• More generally, let α p ∈ Z p (M) be a closed p-form, β p+1 ∈ C p+1 (M) be a

(p+1)-form, z p ∈ Z p (M) be a p-cycle and c p+1 ∈ C p+1 (M) be a (p+1)-chain.
Then D E D E
α + dβ, z p + ∂c p = α, z p .

diffgeom.tex; April 12, 2006; 17:59; p. 185

188 CHAPTER 7. HOMOLOGY THEORY

• Therefore, for every equivalence class [α p ] ∈ H p (M) of closed forms we

can define a linear functional H p (M) → R on the space of homology groups
by, for any [z p ] ∈ H p (M),
D E D E
[α p ], [z p ] = α p , z p .

This is well defined since the right hand side does not depend on the choice
of representatives in the equivalence classes.

• This naturally identifies the space of cycles with the space of cocycles

Z p (M)  Z ∗p (M) .

• For a closed p-form α p and a p-cycle z p the value of the functional

D E Z
α, z p = αp
zp

is called the period of the form α p on the cycle z p .

• We conjecture that
H p (M)  H ∗p (M) .

Proposition 7.4.1 . Let M be a closed manifold. Then for any linear

functional ϕ : H p (M) → R on homology groups there is a closed p-form
• α p such that D E
ϕ(z p ) = α p , z p .
Proof : Difficult.

Corollary 7.4.1 . Let M be a closed manifold. Let k = B p (M) be the p-
p , . . . , z p , be a basis of p-cycles in the homology
th Betti number. Let z(1) (k)

groups H p (M) and π1 , . . . , πk be arbitrary real numbers. Then there is

•
a closed p-form α p such that
D E
p = πi ,
α p , z(i) i = 1, 2, . . . , k .
Proof :


diffgeom.tex; April 12, 2006; 17:59; p. 186

7.4. DE RHAM COHOMOLOGY GROUPS 189

Proposition 7.4.2 . Let M be a closed manifold. Let α p ∈ Z p (M) be a

closed p-form on M such that for any p-cycle z p ∈ Z p (M)
• D E
αp, zp = 0 .

Then the p-form α p is exact.

Proof : Difficult.

Theorem 7.4.1 de Rham Theorem. Let M be a closed manifold. Then
the map
H p (M) → H ∗p (M) ,
that associates to each equivalence class [α p ] of closed p-forms a linear
• functional H p (M) → R on the homology group H p (M) defined by
[α p ] D E
[z p ] 7−→ α p , z p ,

is an isomorphism.
Proof :


7.4.1 Examples
• Torus T 2 .

• Closed Surfaces in R2 .

diffgeom.tex; April 12, 2006; 17:59; p. 187

190 CHAPTER 7. HOMOLOGY THEORY

7.5 Harmonic Forms

• Let (M, g) be a closed oriented n-dimensional Riemannian manifold with a
Riemannian metric g and the Riemannian volume n-form vol .

• Let Λ p (M) be the bundle of p-forms.

• Recall that there is a natural fiber inner product on Λ p defined by

1 i1 j1
hα, βi = g · · · gi p j p αi1 ...i p β j1 ... j p ,
p!
and the corresponding fiber norm
p
||α|| = hα, αi .

• Also, there is a duality between p-forms and (n − p)-forms defined by the

Hodge star operator
∗ : Λ p → Λn−p ,
that maps any p-form α to a (n − p)-form ∗α dual to α defined as follows.

• For each p-form α the form ∗α is the unique (n − p)-form such that for any
p-form β
β ∧ ∗α = hβ, αi vol .

• In components, this means that

1
(∗α)i p+1 ... in = εi1 ... i p i p+1 ...in |g|gi1 j1 · · · gi p j p α j1 ... j p
p
p!
1 1
= p gi j · · · gin jn ε j1 ... j p j p+1 ... jn α j1 ... j p .
p! |g| p+1 p+1

• Recall that the Hodge star maps forms to pseudo-forms and vice-versa.

• Recall also that for any p-form α,

∗2 α = (−1) p(n−p) α ,

meaning that
∗−1 = (−1) p(n−p) ∗ .

diffgeom.tex; April 12, 2006; 17:59; p. 188

7.5. HARMONIC FORMS 191

• The coderivative is a linear map

δ : Λ p → Λ p−1

defined by
δ = ∗−1 d∗ = (−1)(n−p+1)(p−1) ∗ d ∗ .

• The exterior derivative and the coderivative satisfy the important conditions

d2 = δ2 = 0 .

• Problem. Show that in local coordinates the coderivative of a p-form α is

the (p − 1)-form δα with components

1 p 
(δα)i1 ...i p−1 = gi1 j2 · · · gi p−1 j p p ∂ j |g|g jk1 g j2 k2 · · · g j p k p αk1 k2 ...k p
|g|

• Now we define the L2 -inner product of p-forms by

Z Z
(α, β)L2 = α ∧ ∗β = hα, βi vol ,
M M

and the L2 -norm

||α||L2 = (α, α)L2 .
p

• This makes the space C ∞ (Λ p (M)) of smooth p-forms an inner-product vec-

tor space.

• The completion of C ∞ (Λ p (M)) in the L2 -norm gives the Hilbert space

L2 (Λ p (M)) of square-integrable p-forms.

• Let A : H → H be an operator on a Hilbert space H. The adjoint of the

operator A with respect to the inner product of the space H is the operator
A∗ : H → H defined by, for any ϕ, ψ ∈ H,

(Aϕ, ψ) = (ϕ, A∗ ψ) .

diffgeom.tex; April 12, 2006; 17:59; p. 189

192 CHAPTER 7. HOMOLOGY THEORY

Theorem 7.5.1 Let M be a closed orientable Riemannian manifold.

Then the adjoint of the exterior derivative is the negative coderivative

d∗ = −δ .
•
That is, for any p-form α and any (p + 1)-form β,

(dα, β) = −(α, δβ) .

Proof :

1.

Theorem 7.5.2 Let M be a compact orientable Riemannian manifold
with boundary. Then for any p-form α and any (p + 1)-form β,
• Z
(dα, β) + (α, δβ) = α ∧ ∗β .
∂M

Proof :

1. Direct calculation.


• Notice that in case of a manifold with boundary the coderivative is the neg-
ative adjoint of the exterior derivative only on the forms that satisfy one of
the two types of boundary conditions:

α|∂M = 0 or ∗ α|∂M = 0 .

• Let g be the Riemannian metric and ∇ be the corresponding Levi-Civita

connection. It defines a natural connection on the bundle of p-forms. The
Laplacian on p-forms is the operator

∆ : C ∞ (Λ p (M)) → C ∞ (Λ p (M))

defined by
∆ = gi j ∇i ∇ j .

diffgeom.tex; April 12, 2006; 17:59; p. 190

7.5. HARMONIC FORMS 193

• The Hodge Laplacian on p-forms is the operator

L : C ∞ (Λ p (M)) → C ∞ (Λ p (M))

defined by
L = dδ + δd = (d + δ)2 .

• The operators d and δ commute with the Hodge Laplacian, i.e.

dL = Ld , δL = Lδ .

Theorem 7.5.3 For any p there holds

L=∆+W,
•
where W : Λ p → Λ p is an endomorphism on the bundle of p-forms
called the Weitzenböck endomorphism.
Proof :

1.


• Weitzenböck endomorphism is a linear combination of Riemann curvature

tensor, that is, when acting on p-forms W has the form

W i1 ...i p j1 ... j p = F mni1 ...i p kl j1 ... j p Rkl mn ,

where F mni1 ...i p kl j1 ... j p is constructed only from the Kronecker symbol δij and
the metric gi j and gi j .

• Problem. Obtain the expression for the Weitzenböck endomorphism for

p-forms. Hint: replace partial derivatives by covariant derivatives and use
the definition of the curvature.

• A p-form α is called harmonic if

Lα = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 191

194 CHAPTER 7. HOMOLOGY THEORY

Theorem 7.5.4 Let M be a closed Riemannian manifold. Then a p-

• form α is harmonic if and only if it is closed and coclosed, that is,

dα = δα = 0 .
Proof :

1. Use
0 = (Lα, α) = ||dα||2 + ||δα||2 .


• On a closed Riemannian manifold the Laplacian and the Hodge Laplacian

are self-adjoint elliptic operators.

Theorem 7.5.5 Hodge Theorem. Let M be a closed Riemannian

manifold. Then:

1. The vector space H p of harmonic p-forms is finite-dimensional.

•
2. The equation
Lα = ρ
has a solution if and only if ρ is orthogonal to H p .
Proof :

1.

Theorem 7.5.6 Let M be a closed Riemannian manifold. Then any p-
form β is a sum of an exact form dα, a coexact form δγ and a harmonic
form h, that is,
β = dα + δγ + h .
•
In other words, there is an orthogonal decomposition (called the Hodge
decomposition)

Λ p = Im d p−1 + Im δ p+1 + H p .
Proof :

1.

diffgeom.tex; April 12, 2006; 17:59; p. 192

7.5. HARMONIC FORMS 195


Corollary 7.5.1 Let M be a closed Riemannian manifold. Then any
closed p-form β is a sum of an exact form dα and a harmonic form h,
•
that is,
β = dα + h .
Proof :

1.

Corollary 7.5.2 Let M be a closed Riemannian manifold. Then each
de Rham class of cohomologous closed p-forms has a harmonic repre-
sentative.

• Let k = B p (M) be the p-th Betti number. Let z(1)

p , . . . , z p , be a basis of
(k)

real p-cycles in the real homology groups H p (M) and π1 , . . . , πk be ar-

bitrary real numbers. Then there is a unique harmonic p-form h p such
that D E
h p , z(i)
p = πi , i = 1, 2, . . . , k .
Proof :

1.

• The metric g is said to have positive Ricci curvature if its Ricci tensor is
positive-definite.
Corollary 7.5.3 Bochner Theorem Let M be a closed Riemannian
manifold with positive Ricci curvature. Then the first Betti number van-
• ishes, i.e.
B1 (M) = 0 .
That is there are no harmonic 1-forms on M.
Proof :

1. Let h be a harmonic 1-form. Then

Z Z
1
0= ∆ hh, hi vol = Ri j hi h j vol + ||∇ j hi ||2 ≥ 0 .
2 M M

diffgeom.tex; April 12, 2006; 17:59; p. 193

196 CHAPTER 7. HOMOLOGY THEORY

2. Thus h = 0.


• Remark. The elements of the first homology group H1 (M, G) are equiva-
lence classes of 1-cycles.

• The 1-cycles are closed oriented curves (loops) on M.

• If a closed curve can be deformed to a point, then it is a boundary of a

surface (a 2-simplex).

• That is, a closed curve that can be contracted to a point is a trivial 1-cycle.

•
Corollary 7.5.4 For any simply connected manifold M and any Abelian
• group G,
H1 (M, G) = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 194

7.6. RELATIVE HOMOLOGY AND MORSE THEORY 197

7.6 Relative Homology and Morse Theory

7.6.1 Relative Homology
• Let M be a compact Riemannian n-dimensional manifold with smooth bound-
ary ∂M and
i : ∂M → M
be the inclusion map.

• Let x̂i , i = 1, . . . , (n − 1), be the local coordinates on the boundary ∂M and

xµ , µ = 1, . . . , n, be the local coordinates on M in a patch U close to the
boundary. Then the inclusion map is defined locally by

xµ = xµ ( x̂) .

• Close to the boundary there exists a system of local coordinates (xµ ) = ( x̂i , r)
so that the boundary is described by the equation

r = 0.

The coordinate r can be chosen to be the normal geodesic distance from the
boundary so that the vector ∂r is normal to the boundary.

• The inclusion map in this case is given by

xi = x̂i , r = 0.

Therefore,
∂xk ∂r
= δkj , = 0.
∂ x̂ j ∂ x̂ j
• A p-form α on M is called normal to ∂M if

i∗ α = 0 ,

that is,
∂xµ1 ∂xµ p
· · · αµ ...µ (x( x̂)) = 0 .
∂ x̂ j1 ∂ x̂ j p 1 p
• This just means that a normal p-form α vanishes on tangent vectors to the
boundary.

diffgeom.tex; April 12, 2006; 17:59; p. 195

198 CHAPTER 7. HOMOLOGY THEORY

• In other words, a p-form α is normal to the boundary if

α(v1 , . . . , v p ) =0
∂M

for any tangent vectors v1 , . . . , v p .

• In the special coordinate system ( x̂i , r) this means that a p-form is normal if
all purely tangential components vanish on the boundary

α j1 ... j p |∂M = 0 .

• Obviously,
i∗ dr = 0 .
Therefore, a p-form α is normal if there is a (p − 1)-form β such that

α = β ∧ dr .

• A p-form α is called tangent to the boundary if the (n − p)-form ∗α is

normal, that is,
i∗ ∗ α = 0 .

• In other words, a p-form α is tangent to the boundary if

α(v1 , . . . , v p−1 , N) =0
∂M

for any tangent vectors v1 , . . . , v p−1 and a normal vector N, meaning that the
interior product with a normal vector vanishes

iN α = 0.
∂M

• In local coordinates, this condition becomes

∂xλ1 ∂xλn−p µ1 ν1
ελ1 ...λn−p µ1 ...µ p · · · g · · · gµ p ν p αν1 ...ν p = 0.
∂ x̂ j1 ∂ x̂ jn−p ∂M

• In the special coordinate system ( x̂i , r) this just means that the components
with at least one index along the normal direction vanish

α j1 ... j p−1 r ∂M = 0 .

diffgeom.tex; April 12, 2006; 17:59; p. 196

7.6. RELATIVE HOMOLOGY AND MORSE THEORY 199

Proposition 7.6.1 Let M be a compact Riemannian manifold with

smooth boundary. Let α and β be two p-forms on M, which are either
both normal to the boundary or both tangent to the boundary. Then

• (dα, β) = −(α, δβ) .

That is, on normal or tangent p-forms

d∗ = −δ .

Proof : Obvious.


Theorem 7.6.1 Let M be a compact Riemannian manifold with

smooth boundary. Let k = B p (M) be the p-th Betti number. Let
p , . . . , z p , be a basis of real p-cycles in the real homology groups
z(1) (k)

H p (M) and π1 , . . . , πk be arbitrary real numbers. Then there is a unique

tangent harmonic p-form h p such that
•
dh = δh = 0

and D E
p = πi ,
h p , z(i) i = 1, 2, . . . , k .

Proof :

1.

• More generally,

diffgeom.tex; April 12, 2006; 17:59; p. 197

200 CHAPTER 7. HOMOLOGY THEORY

Theorem 7.6.2 Let M be a compact Riemannian manifold with

smooth boundary. Let k = B p (M) be the p-th Betti number. Let
p , . . . , z p , be a basis of real p-cycles in the real homology groups
z(1) (k)

H p (M) and π1 , . . . , πk be arbitrary real numbers. Let γ be a closed

(n − p)-form on ∂M such that for every (n − p)-cycle yn−p on ∂M that is
a boundary of a (n − p + 1)-chain cn−p+1 in M, that is,

i∗ γ = ∂cn−p+1 ,

the value of the integral of γ over y p vanishes,

•
D E Z
γ, y p = γ = 0.
∂M
∂M

Then there is a unique harmonic p-form h p such that

dh = δh = 0 ,

i∗ ∗ h = ∗h|∂M = γ ,
D E
h p , z(i)
p = πi , i = 1, 2, . . . , k .
Proof :

1.


• Let M be a compact manifold with boundary. A relative p-cycle (mod ∂M)

is a p-chain on M whose boundary lies on ∂M.

• A p-cycle with no boundary is called an absolute p-cycle.

• From the point of view of relative homology any p-chain that lies on ∂M is
neglected.

• Example.

• Two relative p-cycles are homologous (mod ∂M) if they differ by a true
boundary and a p-chain that lies on ∂M,

c0p − c p = ∂u p+1 + v p , v ⊂ ∂M .

diffgeom.tex; April 12, 2006; 17:59; p. 198

7.6. RELATIVE HOMOLOGY AND MORSE THEORY 201

• A relative boundary (mod ∂M) is a sum of an absolute boundary and a

chain that lies on ∂M.
• Example.
• The relative homology group is the quotient group of relative cycles mod-
ulo the relative boundaries
H p (M, ∂M; G) = Z p (M, ∂M; G)/B p (M, ∂M; G) .

Theorem 7.6.3 Let M be a compact Riemannian manifold with

smooth boundary. Let k = B p (M) be the p-th Betti number. Let
p , . . . , z p , be a basis of real relative p-cycles of M (mod ∂M) in the
z(1) (k)

real homology groups H p (M, ∂M; R), that is,

k
X
H p (M, ∂M; R) = Rci .
i=1
•
Let π1 , . . . , πk be arbitrary real numbers. Then there is a unique normal
harmonic p-form h p on M such that

dh = δh = 0 ,

and D E
p = πi ,
h p , z(i) i = 1, 2, . . . , k .
Proof :

1.


7.6.2 Morse Theory

• Let M be a closed manifold and f : M → R be a smooth function on M.
• Example.
• A point x0 ∈ M is called a critical point of the function f if
∂f
df = 0, = 0.
x0 ∂xi x0

diffgeom.tex; April 12, 2006; 17:59; p. 199

202 CHAPTER 7. HOMOLOGY THEORY

• A real number a ∈ R is called a critical value of f if the inverse image

f −1 (a) ⊂ M contains at least one critical point.

• A critical point is called inessential if it can be removed by a small defor-

mation of the function f .

• A critical point x0 is called non-degenerate if the Hessian

∂2 f
Hi j =
∂xi ∂x j
is non-degenerate, that is,

det Hi j x0
, 0.

• The Hessian defines a linear operator on the tangent space

H : TxM → TxM

with the matrix

H i j = gik Hk j .

• For a non-degenerate critical point the Hessian operator has non-zero real
eigenvalues.

• The number of the negative eigenvalues (counted with multiplicities), that

is the dimension of the subspace of the tangent space on which the Hessian
is negative-definite, is called the Morse index of the critical point.

Lemma 7.6.1 Morse Lemma. Let M be a closed manifold and f :

M → R be a smooth real-valued function on M. Let x0 be a nondegen-
erate critical point of f with Morse index p. Then in a neighborhood of
•
x0 there exist local coordinates (x1 , . . . , x p , y1 , . . . , yn−p ) such that

f (x, y) = f (x0 ) − (x1 )2 − · · · − (x p )2 + (y1 )2 + · · · + (yn−p )2 .

Proof :

1.


diffgeom.tex; April 12, 2006; 17:59; p. 200

7.6. RELATIVE HOMOLOGY AND MORSE THEORY 203

• The number of critical points of index p is called the p-th Morse type num-
ber and denoted by M p .

• Let t be a formal variable. The polynomial

n
X
M(t) = Mpt p
p=0

is called the Morse polynomial.

• For each real number a ∈ R we define

Ma = {x ∈ M | f (x) ≤ a}

and
Ma− = {x ∈ M | f (x) < a} .

• A real number a is called a homotopically critical value of the function f

if some relative homology group H p (Ma , Ma− ; G) is non-zero.

• It turns out that for non-degenerate critical points a value is homotopically

critical if and only if it is critical.

• That is, for non-degenerate critical points the critical values of f are pre-
cisely the values at which new relative cycles appear.

• Let B p (M) = dim H p (M) be the Betti numbers. The polynomial

n
X
P(t) = Bpt p
p=0

is called the Poincaré polynomial.

diffgeom.tex; April 12, 2006; 17:59; p. 201

204 CHAPTER 7. HOMOLOGY THEORY

Theorem 7.6.4 Morse Theorem. Let M be a closed manifold and

f ; M → R be a smooth function. Suppose that the function f has only
non-degenerate critical points. Let M p be the Morse type numbers, B p
be the Betti numbers, M(t) be the Morse polynomial and P(t) be the
Poincaré polynomial. Then there is a polynomial
n−1
X
• Q(t) = qpt p
p=0

with non-negative coefficients, q p ≥ 0, such that

M(t) − P(t) = (1 + t)Q(t) ,

that is,
M p − B p = q p + q p−1 .

Proof :

1.

diffgeom.tex; April 12, 2006; 17:59; p. 202

7.6. RELATIVE HOMOLOGY AND MORSE THEORY 205

Corollary 7.6.1 There hold:

1. Weak Morse inequalities

Mp ≥ Bp ,

2. In particular, the total number of critical points is bounded below

by the sum of all Betti numbers
n
X n
X
Mp ≥ Bp .
p=0 p=0
•
3. Strong Morse inequalities

M0 ≥ B0
M1 − M0 ≥ B1 − B0
...
Mn − Mn−1 + · · · + (−1)n M0 = Bn − Bn−1 + · · · + (−1)n B0 .

4. In particular,
n
X n
X
(−1) M p =
p
(−1) p B p
p=0 p=0

Proof :

1.


diffgeom.tex; April 12, 2006; 17:59; p. 203

206 CHAPTER 7. HOMOLOGY THEORY

diffgeom.tex; April 12, 2006; 17:59; p. 204

Chapter 8

207
208 CHAPTER 8.

8.1
•

• Theorem 8.1.1
Proof :

1.


1.

diffgeom.tex; April 12, 2006; 17:59; p. 205

Bibliography

[1] T. Frankel, The Geometry of Physics, An Introduction, Cambridge Univer-

sity Press, 1997

[2] M. Nakahara, Geometry, Topology and Physics, Institute of Physics, 2003

[3] B. A. Dubrovin, A. T. Fomenko and S. P. Novikov, Modern Geometry—

Methods and Applications, Parts I and II, Springer, 1984

[4] B. Schutz, Geometrical Methods of Mathematical Physics, Cambridge Uni-

versity Press, 1980

[5] H. Flanders, Differential Forms, Academic Press, 1963

[6] W. M. Boothby, Introduction to Differentiable Manifolds and Riemannian

Geometry, Academic Press, 2003

209
210 Bibliography

diffgeom.tex; April 12, 2006; 17:59; p. 206

Notation

Add all common notation

f :X→Y mapping (function) from X to Y
f (X) range of f
χA characteristic function of the set A
∅ empty set
N set of natural numbers (positive integers)
Z set of integer numbers
Q set of rational numbers
R set of real numbers
R+ set of positive real numbers
C set of complex numbers

211
 212 Notation

diffgeom.tex; April 12, 2006; 17:59; p. 207

View publication stats